In C#/Java memory is not truely released until garbage collection is run and you don't have control over that process. You can give hints, but you don't control when it happens. This is a good thing for most applications since the programmer is no longer required to manage memory which is huge source of bugs.

You can force garbage collection with C# with: System.GC.Collect();
although the call does not mean all resources are reclaimed, it does force the issue instead of waiting for the system to start reclaiming memory.

I've done a CAD like software package and without calling System.GC.Collect(); explicitly, the software would crawl to a halt in 5-10 minutes of heavy CAD drawing.

I've done a CAD like software package and without calling System.GC.Collect(); explicitly, the software would crawl to a halt in 5-10 minutes of heavy CAD drawing.
How did the CAD application finally end up performing? Is it competive with other CAD applications?

How did the CAD application finally end up performing? Is it competive with other CAD applications?

It came out ok for a Visual Studio 2003 C# application, not the fastest on the planet, nor the flashiest, but decent software that was done 2 years ago. (Customer is happy with it, and that is what is important) The software was a custom project for one Customer, they own the rights to the software, it is not available commercially. (They do production related work)

Currently in early phase of version 2.0 which will have a lot of the bells and whistles not implemented in the first version. Trying to find some info on windows Vista, and if it automatically accelerates GDI drawing surfaces, and related functions, if not I guess accessing the GPU would be helpful.

That's what I mean Artmic. You can ask it to reclaim, but it's not full control like you get in other lower level languages. It's actually a great trade off though for not having to manage memory. I did a CAD package a longggg time ago in C. Had some really hard to track down bugs because of memory leaks.

Back to the cell discussion. Thanks for digging up all the great info on the two chips Mterzich. I've enjoyed following along. One of the things you questioned was lack of cache for the SPE's. They don't have it because it's not needed. Unless i'm mistaken, the 256k of local store runs at chip speed. You can read or write in a single clock cycle which is the main reason for cache. I agree with your general conclusions. I would add that I think for a few very specific task the cell will really shine, but as a game processor I'm just not sure it will prove to be a great choice. The 360 appears to be more balanced and approachable for gaming which should mean more time concentrating on features and less time on fighting the hardware.

One of the things you questioned was lack of cache for the SPE's. They don't have it because it's not needed. Unless i'm mistaken, the 256k of local store runs at chip speed.
Normally I would consider the local store with a bus rate of 51.2GB/s to be all that is needed. However, during vector operations, floating point operations are supposed to occur at 8 floating point operations per clock for single precision floating point operations for a total of 25.6 GFLOPS. That seems like lot of data to retrieve and save accross the bus during 1 clock cycle and the amount of data seems to be higher than the 51.2 GB/s bus rate of the SPEs. So is one memory transfer of 128 bits (4 single precision operands) per clock cycle enough data to satisfy the needs of 8 floating point operations per clock cycle? I suspect that I am not understanding something since that rate of data transfer seems extremely high.

The following is the Linpack benchmarks for the cell but it was a 2.4 GHz blade and not the PS3 but it seems to indicate that the cell cores can only maintain a rate of between 8-13 GFLOPS (significantly less on small blocks and less than 5 GFLOPS on blocks below 1024x1024) per core (using 8 SPEs and 1 PPE). The document also states that the 3.2 GHz cell processor was also tested with the benchmarks but did not give the performance improvements expected by the increased clock speed.

http://icl.cs.utk.edu/projectsfiles/iter-ref/pubs/ut_cs_06_580.pdf

http://www.linuxdevices.com/news/NS3591350722.html

From IBM documentation:

The SIMD units on the eight SPEs provide the majority of the computational power of the Cell BE. When using single-precision floating-point fused multiply-add instructions, the eight SPEs in the first-generation Cell BE chip can perform a total of 64 floating-point operations per cycle.
From Linpack Benchmark Documentation:

The SPEs are inherently vector units capable of very fast single precision arithmetic. This work is motivated in particular by the single to double performance ratio of the Cell processor. An SPE can issue a single precision vector fused multiplication addition operation per each clock cycle. A vector of 128B contains four 32-bit single precision values, which means that each SPE can execute 8 operations per cycle.
According to the above quotes, that would seem to keep the bus 100% saturated and able to execute at 25.6 GB/s. However where are the results going? Do the results stay in the vector unit since there doesn't seem to be any bandwidth left to save the results to memory? That is why I am confused.

If the result data is staying in the vector unit, then that result would need to be used as a source operand for the next multiply addition operation producing more a total of all values in the vector instead of the more conventional vector math producing individual results for each operation.

For the life of me I can't fiqure out a real life situation where a total of all elements in an array would require repetive multiply-additions or multiply-subtractions to produce one total of all the values in the array. But then I am not a mathematician or a physicist.

On my previous calculations, I always assumed that results must be sent back to memory so I always came up short in GFLOPS. However, if the results can remain in the vector unit, the maximum GFLOPS should be as stated by Sony and IBM at 25.6 GFLOPS.

What Carmcak really said was

They are both powerful systems that are going to make excellent game platforms, but I have a bit of a preference for the 360’s symmetric CPU architecture and excellent development tools," he said. "The PS3 will have a bit more peak power, but it will be easier to exploit the available power on the 360. Our next major title is being focused towards simultaneous release on 360, PS3, and PC."


I think the real answer is that PS3 is theoretically more powerful. However, it remains to be seen whether that extra power will ever make a difference in any game because it is difficult to tap.

Bottom line, I am really leaning towaard the Xbox. First, it is available. Second, the games right now are better. Third, I can get hd-dvd right now. If bluray catches on, I am sure an addon will be made available.

Although Sony and other manufacturers like to tout the peak GFLOPS of their systems, when running the Linpack benchmarks they all seem to underperform. Whether the xenon or cell processor can produce 2 or 25 GFLOPS per core in actual implementations, although interesting, will not make a significant difference as to whether either console will be a success.

Finding and segmenting code that can fit in an SPE memory can be a daunting and difficult task. But if the code cannot execute in parallel, the effort is in vain. Code like everything in life has dependencies but in life you are performing many unrelated tasks so it is easy to find parallelism allowing jobs to be completed more quickly if several people were available to perform those tasks.

However, a game application basically only performs one large task that gathers control information from the user and paints the screen. In a game application, there is only one obvious major case of possible parallelism. The graphics of a screen cannot be developed and painted until the information from the AI is generated but once the AI information is available, the graphics can be produced while the AI is generating the information for the following screen creating parallelism with a one frame latency.

Since the xenon processor can easily implement this characteristic of the application, immediate results should be seen. However, with the cell processor due to the restricted memory size of the SPE, capturing the power of this concept will be very difficult.

If a large amount of parallelism can be created with the cell processor, the cell processor should perform better than the xenon processor. The most important question is whether that power can be harnessed.

From all appearances, it appears that Microsoft is well in control of its destiny as far as game consoles are concerned.

True next generation game console
1 year head start over PS3
Basic console is less expensive than PS3
Exclusive applications are currently available
Relatively easy to program
Very good online gaming
Doesn't loose money on the sale of each console

Although Nintendo is a major competitor with the Wii and probably will be in the near future due to its price advantage as well as there are still limited numbers of HDTVs, but several years down the line that game console will probably become obsolete.

IMO, Microsoft could go after Sonys juggler by producing a console that would be a fully integrated economical to produce console for Christmas 2007.

A 2.5 GHz 3 core PowerPC with out-of-order execution giving performance improvement of more than 50%
Faster GPU
Internal power supply
Optional builtin HD DVD or DVD
Optional 120 GB hard drive
Quieter operation

The cost of such a system (maybe called Xbox 360+) should probably not be significantly more expensive since the processor clock rate is reduced but should perform better than the current console. Also since the processor has exactly the same instruction set, emulation of current applications would not be necessary (such as what was done for the original xbox games), and those games may perform better. Development tools and compilers would remain the same.

The biggest issue is the ability to have exclusive killer applications on the 360 game console. It is important that Microsoft sign up as many exclussive titles that is possible and economical. Nintendo in the furture (also Sony if the cell does not work out) could design a standard 3-core system and have any non-exclusive title easily ported to that console (probably well less than 5% of the development cost per application ported). Currently it is not as important that Microsoft locks in killer applications that require more than one core since the Wii does not have the power or hardware to run the application and Sony has such a non-standard processor that porting to that console will be difficult, time consuming, and expensive. But if a competitor produces a standard 3-core console in the furtue, that manufacturer could quickly catch up with inexpensive ports if the titles are not exclusive.

In the past, Sony had an easy time getting developers to produce exclusive titles for the PS2. It is my understanding that console manuufactures of last generation game consoles had to give some incentive to game developers to create exclusive tiltles such as the following.

Loan to finance game development to be paid of after the game is released
Low or interest free loans
Loan may not have to be paid back if the game is a flop
Outright grant to develop games
Reduced royality payments

Since Sony had a major share of the market, I suspect that Sony usually only had to only give a loan at current market interest to get developers to create exclusive titles. However Microsoft probably had to make many more concessions to get developers to produce titles for the original xbox. It now appears that the tide has turned in favor of Microsoft away from Sony due to being first to market and easier to program.

It appears that Sony as well as the game developers realized that it was going to be much more difficult and expensive to develop games for the PS3. With exclusive game development for the Xbox 360 expected to cost significantly more than previous generation games and development for the PS3 is expected to be significantly more expensive than the 360, exclusive developers for the PS3 were concerned due late to market, higher cost of development, larger sales requirement, and higher risk.

It appears that Sony developed a plan back in Feburary for Korean developers to try to allievate some of the problems but developers were less than satified. The following are some of the highlights.

Sony will provide the entire development cost for an online PS3 title. The project will be funded up to the Alpha version
Even if the prototype of the game is finished, the quality or marketability could be deemed "low" and the entire development cost would need to be paid back to Sony
The entire rights including the source code would be owned by Sony regardless of whether the title is published or not.
If the game is successful and a sequel is made, Sony will own all rights to the title.
When the game is released, all initial profit will go to the Sony to cover the development cost. After that is paid off, Sony and the development team would divide the remaining profit.

I would doubt that developers would agree with Sonys plan.

http://www.kotaku.com/gaming/korea/sony-pisses-off-online-korean-developers-155838.php

There is no chance for anything like a XBOX 360+. MS would not make a refresh of a current console that had differing hardware permutations. This would cause an uproar with their current customers and also further raise costs in developing, testing, and designing new games when you have 2 differing systems. There is no chance for anything like that to happen.

There is no chance for anything like a XBOX 360+. MS would not make a refresh of a current console that had differing hardware permutations. This would cause an uproar with their current customers and also further raise costs in developing, testing, and designing new games when you have 2 differing systems. There is no chance for anything like that to happen.
The hardware is the same but just faster, possibly more capacity, using the same operating system so there shouldn't be any added development or testing costs. This design would possibly improve any potiential frame rate issues and potiential stalls that would be in the original 360. I agree that some customers may feel left behind.

It would probably be very inexpensive to design and produce such a system so I think Microsoft could be considering such a system.

I agree with Hmerly.

There's no way any console manufacture would "improve" their console like this. When that happens, they system will no longer be a closed box system (so do developers design on the newer model or the older model? If they design a game based on the newer model it would render the game essentially unplayable on the older model).

To do so is suicide.

The hardware is the same but just faster, possibly more capacity, using the same operating system so there shouldn't be any added development or testing costs. This design would possibly improve any potiential frame rate issues and potiential stalls that would be in the original 360. I agree that some customers may feel left behind.

It would probably be very inexpensive to design and produce such a system so I think Microsoft could be considering such a system.

I think some would be an understatement. To release a more powerful version of the same console would be akin to releasing a SegaCD, 32X, etc., except you'd have to buy an entirely "new" console. I can't see such a move as being anything less than a slap in the face of current customers.

The hardware is the same but just faster, possibly more capacity, using the same operating system so there shouldn't be any added development or testing costs. This design would possibly improve any potiential frame rate issues and potiential stalls that would be in the original 360. I agree that some customers may feel left behind.

It would probably be very inexpensive to design and produce such a system so I think Microsoft could be considering such a system.

No chance of anything like your 360+. The costs would be significant (MSalready paid IBM and ATI to develop the current IP, redesigns for faster speeds will cost), the development/QA testing headaches would be doubled and you'd piss of most of your current customers. All so you could fulfill a small niche markets dream? The next xbox will be released in 4-5 years, anything less doesn't justify the money they invested in this gen. Especially with the PS3 struggling.

BTW its much better to spend the money enticing developers to come to the 360 and working on lowering the component prices so you can lower console prices. The number of sales MS can get by dropping the core system into the $200 range is huge. Especially of most of the AAA titles are coming out on both systems.

OK, you guys have convinced me that increasing the processor power is not a good idea. I was looking at it from a marketing point of view and not from a technical point of view. I assumed that developers would still develop the applications for the original 360 and benefits for the newer model would be marginal. However, if some developers developed for the newer model causing the older model to not work very well, that could end up being a marketing disaster for Microsoft.

Perception many times is different from reality. I initially was assumng that Microsoft could tout an upgraded game console that is 50% faster but in reality the power would never be used. However, as you guys pointed out, the possible dangers outweigh any possible benefits.

I think some already perceive that the 360 is not as good of a game console as the PS3 except for most of the gamers that already own the 360 or both consoles. Even though many that own both consoles believe that the 360 is currently better for gaming, some believe that the PS3 will eventually catch up and eventually surpass the 360. If Microsoft sits on it laurels too long, opportunities can pass it by. Even if they have a superior console, that does not guarantee success. Perception is important. This is why gamers are are purchasing the PS3 with the belief that games will get better but it is anyones guess when or if that will occur.

Technical perception is important to a degree, but ultimately the exclusive titles are what matters, otherwise PS2 should've been crushed by the original Xbox.

Right now most gamers would concur that the 360 and PS3 are at least comparable in power. Even if one ends up looking better, it's not like the other one will suddenly look like a N64. So the deciding factor is still the availability in games.

Right now most gamers would concur that the 360 and PS3 are at least comparable in power.
Is that really true? I don't own either but from what I can gather the PS3 sometimes has lower frame rates and can stall more often using the same application. Is that a processor power issue or a GPU issue? Are people currently giving Sony the benefit of the doubt assuming that is due to it being a ported application?

Is that really true? I don't own either but from what I can gather the PS3 sometimes has lower frame rates and can stall more often using the same application. Is that a processor power issue or a GPU issue? Are people currently giving Sony the benefit of the doubt assuming that is due to it being a ported application?

I think that is true. For every person who points out that PS3's launch line up performs worse than their 360 counterpart, another person will say "give it time and PS3 will pass the 360". So if you average the two camps out, you can probably make the assessment that people believe they are comparable.

I think it's relatively easy to assume why the PS3 multiplatform games aren't doing so well right now.
-They are probably running on the PPE only with next to no SPE usage (actually I doubt most of those games use all 3 cores on the 360 as well...Tony Hawk/CoD3? Come on...)
-360's GPU gives it the edge over RSX, and that's power easily harness-able.
-PS3's harddrive also has not been utilized in many games (Fight Night 3), and the slower data transfer speed of the Blu Ray also results in longer loading time than 360.

All those factors combined together are what's making Sony looking bad I think.

Technical perception is important to a degree, but ultimately the exclusive titles are what matters, otherwise PS2 should've been crushed by the original Xbox.
Yes, technical perception isn't the most important factor, but I seriously doubt that most gamers even considered the original Xbox superior to the PS2 (from a merely technical point a view, not taking the software into account).
That's why I don't find it surprsising at all that most people will gladly believe that PS3 is 2x more powerful than 360.

Right now most gamers would concur that the 360 and PS3 are at least comparable in power. Even if one ends up looking better, it's not like the other one will suddenly look like a N64. So the deciding factor is still the availability in games.
If by gamers you mean "hardcore" gamers, that try to keep themselves the least bit informed, then probably yes; but I'm willing to bet the vast majority of people don't have the slightest doubt in their minds that PS3 is vastly superior, and the only thing that will make them consider the 360 as a possible purchase (aside from the shortages, obviously) really is the price point.

I'm willing to bet the vast majority of people don't have the slightest doubt in their minds that PS3 is vastly superior, and the only thing that will make them consider the 360 as a possible purchase (aside from the shortages, obviously) really is the price point.
Very well put. Microsoft has a small Window to try to change that perception even if it is all smoke and mirrors.

the only change MS will be doing to the design is going to a 65nm manufacturing process for the chips early next year. and hopefully a redesign of the internals to make it run cooler (which inherently happens with a 65nm process) and work out the bugs causing 3 red ring death and random DRE's

this is typical of all consoles, but the hardware specs will remain unchanged. this is a console not a pc.

but I seriously doubt that most gamers even considered the original Xbox superior to the PS2 (from a merely technical point a view, not taking the software into account).

I think I would have to disagree with that; I have never came across a gamer that believes PS2 had better graphics than the Xbox. Xbox had noticeably better graphics since its launch with Halo and Doa3, not to mention EVERY SINGLE multi-platform game looked better on Xbox. The difference is too drastic - even the most hardcore PS2 fans on forums admit Xbox had better hardware.

It seems pretty easy to understand how a game application can be developed to use two general purpose cores in parallel. Usually the main problem is finding code that can execute in parallel and that all the needed data for both cores is available to be processed when both cores are ready to execute. In the case of a game application the graphics cannot not be developed until the AI has been produced since the graphics code needs that data produced by the AI code. However once the AI is produced, the graphics can then be developed and displayed for one frame while the AI code is developing the AI for the next frame creating the parallelism.

It is also pretty easy to see how massively parallel processing works in a physics environment. In this case the main processor can load programs into each of the other cores and than notify each of the cores to perform certain calculations using certain data and send the results to certain places. When a core finishes calculating the results, the main processor will restart that core indicating a new calculation but using different data. The main processor would need to manage the whole process as well as put together all the different results to be able to come up with the overall final desired results.

However, how is the game developer going to be able to use the cell processor with the limited memory capability for the SPEs? Will it be performed like the massively parallel processing as done for physics? Although a game application uses some physics code, is the data going to be available at the time needed to be able to execute the code in parallel? Is there enough physics code available that can be executed in parallel to keep the SPEs busy or will it only require a few percent of the SPE processing power?

If the developer finds that he cannot keep the SPEs busy executing only physics code will he then attempt to break the code down into 100s of segements that will be paged into the SPE memory? Since a game program can be as large as 256 MB including the needed data, will the developer try to break that program into 500-1000 segments and page that code and data into SPEs? If 500 segments was used, that would require 30,000 pages per second to occur. Would the paging of that many pages and the managing of the code and data be counter productive and use more processor power than could possibly be gained?

I think I would have to disagree with that; I have never came across a gamer that believes PS2 had better graphics than the Xbox. Xbox had noticeably better graphics since its launch with Halo and Doa3, not to mention EVERY SINGLE multi-platform game looked better on Xbox. The difference is too drastic - even the most hardcore PS2 fans on forums admit Xbox had better hardware.
Well, I could be wrong, but I'm still not convinced.
Like you said, even PS2 fanboys on forums admit it, because they probably have seen what both consoles are capable of, they compared the multi-platforms titles, in short, they are reasonably aware of what each machine is capable of.

However, I can't imagine that the bulk of the 100+ million PS2 owners have actually cared to compare the options available before purchasing their console. They probably buyed it because their friends already had one, or because they saw an ad for GTA and thought it was a PS2 exclusive, because the store displayed PS2 titles a lot more proeminently than other consoles' and they thought it meant they were better, or just because it was a "PlayStation", plain and simple.

I'm not implying that the sales numbers would be drastically different if every consumer spent at least a few weeks considering their purchase, there are loads of plausible reasons for picking up a PS2 instead of an Xbox/GameCube and vice-versa, but I just don't believe that a significant proportion of PS2's user base (or any other console, for that matter) went through the hassle of thouroughly comparing the 3 options.

Nice questions mterzich! I guess only time will tell once a PS3 developers gets accustomed in coding for the cell. However, perhaps we can find a developer that has dabbled with cell?

I've been wondering why the PS3 sets aside a SPE for the operating system since it appears that a SPE would be rather useless for that purpose. I suspect that it is not really being used for the operating system but instead Sony has intentions to use a SPE for library support.

Since the SPEs are quite difficult to program, I suspect that Sony is attempting to use an SPE to semi-automatically support some of the more time consuming library calls. If Sony can find several time consuming library routines that fit into the SPE memory, it can switch to the SPE to execute those library calls.

The only requirement for the developer to use the capability would be to have the AI and the Graphics executing on different threads. When one of those SPE supported library calls would be made, the thread that made the call would be suspended giving 100% of the PPE processor power to the other thread. The SPE would then be running in parallel with the remaining thread effectively doubling the processor power while it was executing.

A page swap mechanism to swap different library pages into the SPE could also be implemented by Sony but if 200 KB worth of code and was swapped, only routines that could save the equivalent of at least 15,000 assembly language instructions could be used in the SPE otherwise it could possibly have a negative performance effect.

I've seen some of things the CELL is capable of already that I haven't seen in any 360 or PS3 title as of now, like dynamic weather effects, i.e., the snow effect in Resistance and the rain/cloud in F1. As far as animation goes, Naughty Dog's new jungle game, aka BIG, seems to trump all games I've seen that have been announced so far. Factor 5 have said that Lair couldn't be done on the 360 due to the amount of characters having complex A.l., post processing effects, and the range of scope per square feet. Back to F1, the game sports MANY cars at onces with advance physics and the volumetric lighting is amazing.

EDIT-Here's the quote from the IGN interview:

Eggebrecht: Lair in its current form couldn't be done on 360. We are using large amounts of Cell's SPUs for all of our geometry, landscape, simulations, animations, even troop AI.

I agree that the cell processor has the potiental to shine compared to the xenon processor.

If we assume that a xenon PPE and a cell SPE has similar performance capabilities and all 6 SPEs were executing approximately 33% of the time, the same performance can be done with the xenon processor by using one PPE core for the main program and the other 2 cores to execute the AI code using 100% of the 2 core processor power.

If the code executed by the SPE is primarily vector math and if we believe the specifications, then each xenon PPE has a peak performance of 38 GFLOPS where as each SPE only has 25.6 GFLOPS. In that case, it would require all SPEs to execute more than 50% of the time to exceed the processor capability of the xenon processor.

However, if all 6 SPE cores can efficiently execute at or near 100%, the cell should well outperform the xenon processor.

However, if all 6 SPE cores can efficiently execute at or near 100%, the cell should well outperform the xenon processor.

Which is what Factor 5 seems to be doing.

The thing is, it isn't going to happen for the most part. Like I've said in other threads already, it's going to be like the PS2 where you'll only have a handful of developers that'll be able to tap the power of the console, like Konami, SquareEnix, Capcom, Namco, Sony's first/second party developers and I'm sure I'm missing a few others that I can't think of right now.

This thread is fantastic. Good job mterzich for doing all the research and analyzing of the datat that you have found out. As somebody suggested you should seriously think about getting that information into a more Coherent form on a Website or Blog.

I am very tempted to even to it myself.

Only time will tell. Currently there doesn't appear to be much available for either the 360 or the PS3 to indicate the full capability of either processor.

The PS2 had a different problem than the PS3 in that Sony delivered a toolkit that didn't work very well. Developers were able to get around most of the problems (if the programming language was C) by using a tookit for PC games and initially developing and debugging the application for the PC since the PS2 had a standard archeticture. The last 5% of the development would then done with the PS2 tookit. If the supported language was assembly language that would be a little more difficult but that could also be developed using the PC and when completely developed and debugged, that source could be recomplied using a MIPS complier to produce and assembly language listing which could then be used as the source for the PS2 that could be modified. It is my understanding is that the toolkit delivered for the PS3 was written by a third party, supports several languages, and works fairly well.

From everything that I have read, the AI and graphics generation each required about the same amount of processor power during the last generation game applications. If that remains the same for this generation games, even if 100% of the AI code is off loaded to the SPEs, the PS3 would not be able to execute any faster than a 2-core PPE processor since the limiting factor would be the speed of the graphics generation. However, if the AI code requires 6x the processor power of the graphics generation code and all the AI code was off loaded to the SPEs, than it can perform at 100% or about 2x or more the speed of the xenon processor. With between 2x and 3x the ratio, the cell and the xenon should perform similarily if both were implemented efficiently.

If it is less than the ideal 6 to 1 ratio of AI to graphics power requirement, than something would need to be done to off load the graphics generation to SPEs to utilize all the processor power of the cell processor. This is not an easy task since the graphics library is usually very large (many graphics libraries are more than 30 MB but I don't know the size of the Sony library).

The GPU will only have an effect on performance of either processor if it cannot process frames at the rate that the frames can be produced (maximum 60 fps at 720p or 1080p or 30 fps at 1080i). If that occurs, than either processor will slow to the maximum frame rate that the GPU can process. Both the AI and graphics generation can usually only process the next frame in advance while the GPU is processing a frame. If more than that is processed in advanced, noticable lag will probably start to occur. So neither processor will perform very well if the GPU is bogged down processing frames.

I stated in the last post that the GPU can have an effect on performance of the processors if it cannot process frames as fast as can be generated by the processor. Although that is true there are exceptions, some applications disable VSync giving the appearance that frame rates are higher than they should be but introduce tearing caused by frames possibly being intermixed. The only advantage of this concept is that more frames can be issued but it creates something less than desireable on the screen.

Also I stated that more than one frame should not be processed in advance due to the possibility of lag. However, DirectX allows for up to 3 frames to be buffered if desired (not sure about the libraries for the cell or xenon processors).

The following are extracts from the NVIDIA Programming Guide.

Enabling VSync forces the frame rate to (at best) match the monitor refresh rate. Multi-GPU configurations are likely to achieve frame rates that are much higher than monitor refresh rates, due to their higher graphics performance. These frame rates are thus only attainable if VSync is turned off.

For some applications, i.e., visual simulations, VSync is an absolute requirement to avoid tearing. In these cases SLI still demonstrates significant benefit when the graphics workload and fill-rate requirement are high. For example, using SLI it may become possible to run those applications in resolutions of 1920×1200 or 2048×1536 along with 4× or higher full-scene antialiasing at real-time rates.
DirectX allows drivers to buffer up to three frames. Buffering frames is desirable to ensure that CPU and GPU can work independent of one another and thus achieve maximum performance. On the other hand, the more frames are buffered the longer it takes between issuing a command and seeing its result on-screen. This lag is generally undesirable, since humans can detect and object to lag-times of as little as 30 ms (depending on the test scenario).

Some games thus artificially limit how many frames are buffered. For example, locking the back-buffer forces a hard synchronization between the CPU and the GPU. Locking the back-buffer first stalls the CPU, drains all buffers, and then stalls the GPU. At the end of the lock all systems are idle, and the number of buffered frames is zero.

Stalling a system in this manner has a severe performance penalty and is to be avoided, especially on multi-GPU configurations.
A less objectionable solution to limiting the number of buffered frames is to insert tokens into the command stream (e.g., DirectX event queries) at the end of each frame. Checking that these events have been consumed before issuing additional rendering commands limits the number of buffered frames and thus lag to anywhere between one to three frames.
http://developer.download.nvidia.com/GPU_Programming_Guide/GPU_Programming_Guide.pdf

The GPU should be able to process each frame in less than 1/60th of a second for 720p or 1080p or 1/30th of a second for 1080i if it is desired to have the maximum frame rate and no tearing. If the GPU cannot perform at that capability, the frame rates could possibly decrease by as much as half since they could miss the v-sync and have two identical frames displayed. Other than reducing the complexity of the screen, the developer can probably only try to off load more graphics generation to the system processor to reduce the strain on the GPU.

In an earlier post in this thread, I speculated on how the SPEs in the cell processor will be managed. I wondered if the application would use a paging system or possibly create resident code for each SPE. The following article may put a little more light on that subject. Several proposals have been made and development is being performed for tools to use some of those different options.

http://www.blachford.info/computer/Cell/Cell3_v2.html

The cell is way more powerful than the xenon. It's not even close.

Cell SPU can access any of its memory in one single instruction, but the Xenon must first load an address followed by a load instruction, meaning Xenon typically needs two cycles per vector load.

A hinted branch on SPU takes 1 cycle, unhinted 15 cycles.
A hinted branch on Xenon takes 8 cycles, unhinted 24+ cycles.

SPU has efficient conditional selection instructions so you can eliminate many branches from code. Xenon does not, meaning more branches (8+ cycles each).

Cell has access to 36GB/s memory. Xenon has access to 22GB/s memory, which is shared with the GPU.

Cell SPUs can issue 16 parallel asynchronous DMA memory transfers between SPUs and memory. Xenon has no such DMA.

Cell SPUs have a total of 7*256 = 1,792KB local memory (eqv to lv1 cache). Xenon has just 3 *32 = 96KB lv1 cache.

The 360 has lower memory bandwidth and a shared memory architecture so bandwidth is shared between CPU and GPU. So while the GPU is chewing up textures the rest of the system will suffer.

The PS3's XDR memory is specially designed for high performance data streaming to the SPUs. It is fast enough that each SPU can function at maximum capacity before bandwidth becomes any kind of problem. The SPUs can also interchange data without touching XDR at all for even higher bandwidth.

The PS3 GPU memory bandwidth is not a shared memory system. GPU has it's own high speed DDR3 ram. If a game developer wanted they can make use of XDR + DDR3 ram in parallel, dwarfing the 360 by far.

Just a few of many factors why the Cell/ps3 can sustain much higher performance than Xenon/360.

A hinted branch on SPU takes 1 cycle, unhinted 15 cycles.
A hinted branch on Xenon takes 8 cycles, unhinted 24+ cycles.


The Xenon is pipelined. So while it may take 8 cycles from instruction fetch to instruction retire, there is still other work getting done in those 8 cycles. Basic computer architecture stuff.

In an earlier post in this thread, I speculated on how the SPEs in the cell processor will be managed. I wondered if the application would use a paging system or possibly create resident code for each SPE. The following article may put a little more light on that subject. Several proposals have been made and development is being performed for tools to use some of those different options.

http://www.blachford.info/computer/Cell/Cell3_v2.html

Positive contributions to a great thread, with the wisdom to recognize that at this time, the discussion is entirely.....academic. Real-world efforts in the development community will eventually segregate rash pronouncements of clear, obvious on-paper technical superiority from more conservative, thoughtful speculation about each systems' ultimate performance potential....and practical hardware limitations. Just a wild guess straight out of my ass here, but I wonder if at the point when developers have come to grips with the vagaries and idiosyncracies of both systems, any theoretical presentation advantages one machine might have over the other will be no more, or perhaps, less apparent in this generation, than they were in the last.

Cell SPU can access any of its memory in one single instruction, but the Xenon must first load an address followed by a load instruction, meaning Xenon typically needs two cycles per vector load.
I don't know what you are refering to but the PPE address is part of the load instruction just like the SPE. Both are required to translate the load instruction to determine its type and then cause the load of the data from the designated address. Since IBM does not publish the speed of L1 cache of any PowerPC, I will quote the Apple G5 documentation.

L2 Cache
512K of L2 cache provides the core with ultrafast 64GBps access to data and instructions.

L1 Cache
Instructions are prefetched from the L2 cache into a large, direct-mapped 64K L1 cache at 64GBps. In addition, 32K of L1 data cache can prefetch up to eight active data streams simultaneously.

Fetch and Decode
As they are accessed from the L1 cache, up to eight instructions per clock cycle are fetched, decoded and divided into smaller, easier-to-schedule operations. This efficient preparation maximizes processing speed as instructions are dispatched into the execution core and data is loaded into the large number of registers behind the functional units.If the specifications are anywhere near the same, it is highly unlikely that under most cases that an SPE can execute a load instruction any faster than a PPE.

A hinted branch on SPU takes 1 cycle, unhinted 15 cycles.
A hinted branch on Xenon takes 8 cycles, unhinted 24+ cycles.A hinted branch on the xenon processor does not normally take any longer than a hinted branch on a SPE since the xenon L1 cache and the SPE local store are both very fast. In the case of the xenon processor, the hint will trigger the L1 cache in background to transfer the data block containing that instruction to the cores execution pipeline. Even if L1 cache is missed, it is very possible that the hinted block is in the L2 cache and even though there is a latency of between 5-15 clocks for that block in the L2 cache, the processor is piplined and the branch hint is performed well in advance allowing in most cases no stall at all. A miss of L2 cache can cause a processor stall of between 100-300 clocks due to latency of the main memory. Transfers from main memory to the L2 cache to the L1 cache to the instruction pipeline is performed in blocks.

Depending on the size of the pipeline (number of non-contigious blocks it can contain) as well as forward looking capability of the branch predictor will determine if and when the processor will stall. Since the size of the pipeline is not documented for either the SPE or the xenon processor or how far and sophisticated the branch prediction capability is implemented in the xenon processor, it cannot be determined which is suited better for branch prediction. Although I suspect that the xenon is better suited for branch prediction otherwise Sony would have requested that the branch prediction capabilities in PPE core be removed to save on cost and just use software hints like the SPEs. Microsoft would also have had the branch prediction capabilities completely removed in the the xenon cores to save cost and would also use software hints just like the SPEs of the cell processor.

SPU has efficient conditional selection instructions so you can eliminate many branches from code. Xenon does not, meaning more branches (8+ cycles each).Generally this thread has always pointed to documents when statements are made. Can you point to such documents.

Cell has access to 36GB/s memory. Xenon has access to 22GB/s memory, which is shared with the GPU.Where did you get the 36 GB/s memory speed for the cell? According to all documents from IBM, the PPEs main memory speed is 25.6 GB/s (64 bits per clock) and the SPE is memory speed is 51.2 GB/s (128 bits per clock).

Cell SPUs can issue 16 parallel asynchronous DMA memory transfers between SPUs and memory. Xenon has no such DMA.I agree but why would you want to do such a thing with a 3-core shared memory processor. DMAing between memories is extra overhead.

Cell SPUs have a total of 7*256 = 1,792KB local memory (eqv to lv1 cache). Xenon has just 3 *32 = 96KB lv1 cache.
Yes, however the xenon also has 1 MB of shared L2 cache which under many cases will not stall the processor. If it does, other theads on the core will usually just keep on working but at a faster pace. That is what parallel processing is about.

The 360 has lower memory bandwidth and a shared memory architecture so bandwidth is shared between CPU and GPU. So while the GPU is chewing up textures the rest of the system will suffer. The PS3 GPU memory bandwidth is not a shared memory system. GPU has it's own high speed DDR3 ram. If a game developer wanted they can make use of XDR + DDR3 ram in parallel, dwarfing the 360 by far.There may be some truth to that. However, the xenon processor and the 360 GPU has the ability to directly use the L2 cache for procedural synthesis and according to most reports, the 360 GPU performs better than the PS3 GPU. I suspect that it is currently too early to determine which concept will perform better.

The PS3's XDR memory is specially designed for high performance data streaming to the SPUs. It is fast enough that each SPU can function at maximum capacity before bandwidth becomes any kind of problem. The SPUs can also interchange data without touching XDR at all for even higher bandwidth.Although the each SPE can transfer data at a 51.2 GB/s rate to and from the EIB, if they are all transferring data to and from main memory, the maximum transfer rate for all SPEs combined cannot exceed the maximum data rate of the main memory bus which is 25.6 MB/s. Since this transfer rate is only about 8% more than the GDDR3 memory bus in the 360, most of that and possibly more can be made up by the 360 in that cache automatically triggers block reads and writes to and from main memory. The cell has to program the DMA and then wait for it to transfer before it can use the memory. Of course, the cell can try to create sophisicated routines to create ping-pong effects or try to use only the memory that is available constantly checking the DMA device to make sure the memory is available prior to using it to try to eliminate the latency issue.

No one is debating that the cell has a high degree of potiential. Until future applications are written, it cannot be determined if that potiential can be harnessed and at the same time be cost effective.

I've seen some of things the CELL is capable of already that I haven't seen in any 360 or PS3 title as of now, like dynamic weather effects, i.e., the snow effect in Resistance and the rain/cloud in F1. As far as animation goes, Naughty Dog's new jungle game, aka BIG, seems to trump all games I've seen that have been announced so far. Factor 5 have said that Lair couldn't be done on the 360 due to the amount of characters having complex A.l., post processing effects, and the range of scope per square feet. Back to F1, the game sports MANY cars at onces with advance physics and the volumetric lighting is amazing.

EDIT-Here's the quote from the IGN interview:

I don't know what to believe. It mostly subjective don't you think? I don't know if Factor 5 is telling the truth or helping PS3 sales because they are developing on it.

Define dynamic effects please.

Well this here is interesting to read from a Phil Harrison:

http://www.1up.com/do/newsStory?cId=3155972

Sony: No One Will Use 100% of PS3
Says their games currently use less than half.
by Patrick Klepek, 12/18/2006
57 of 57 users recommend this story.

Speaking to MTV, he said "nobody will ever use 100 percent of its capability" due to firmware upgrades and new uses for the Sixaxis controller.


.

After reading through your posts about the SPE's mtez. It might be assumed that this could mean that basically the SPEs just aren't capapble enough or that game developers just won't be able to code properly to make use of the SPE's fully?

What is your take on this from your academic and technical understanding as one generally don't issue such a statment unless you are trying to cover up some sort of problem?


EDITED:
Didn't want to bring to much of the "Console" War issue into this thread as it is a good thread, but the quote from a top Sony Exec shouldn't be ignored.

Positive contributions to a great thread, with the wisdom to recognize that at this time, the discussion is entirely.....academic. Real-world efforts in the development community will eventually segregate rash pronouncements of clear, obvious on-paper technical superiority from more conservative, thoughtful speculation about each systems' ultimate performance potential....and practical hardware limitations. Just a wild guess straight out of my ass here, but I wonder if at the point when developers have come to grips with the vagaries and idiosyncracies of both systems, any theoretical presentation advantages one machine might have over the other will be no more, or perhaps, less apparent in this generation, than they were in the last.
It is purely academic. The only facts are that the cell processor is a single chip revolutionary processor for use in game development with some difficulties in the ability for the developers to develop game applications for that processor. The xenon processor is just an evolutuionary processor for game development.

If we only had discussions about the hardware archeticture and speed of both processors and the potiental of each without considering any of the issues with developing game applications, this would not reflect how the hardware could be used and the thread should have been closed at the first post.

That is why I didn't dwell on whether one has faster memory, which has a better GPU, overhead of the DMA, lack of branch prediction, extra overhead of DMA, number of cores available, number of busses availabe, lack of out-of-order execution units, etc. Discussions of those capabilities were to allow people to understand how the system was put together. Although all of those are important, they are relatively unimportant to the developer in most cases compared to the small memory size of the SPE.

Everything else is pure speculation. But from my own programming experiences it seemed appropriate to indicate issues that developers may have developing game applications using SPEs.

The primary questions about the PS3 would probably be as follows.

Is the cell a very powerful processor? Yes
If the cell and the xenon processor were both programmed to use all available resources, will the cell outperform the xenon processor? Most probably
Will the SPEs get easier to use as more and better toolkits are avalable? Yes
Will developers be able to become more proficient developing code for the SPE as they gain experience? Absolutely
Will developers enjoy developing code using SPEs? Probably initially but probably not after they did it once since it would probably be considered dirty work
Will development using SPEs ever be anywhere near as easy as developing for the xenon processor? Only if someone develops a virtual operating system to automatically break the code and data into segments that would fit in the SPEs and would page that code and data in and out of SPEs memory
Because of the PS3s theoritical power advantage, will programmers be inclined to spend the extra time required to develop superior applications for the PS3? I don't know
Will the PS3 be a success? Just like Iraq. What is the definition of success?

I don't know what to believe. It mostly subjective don't you think? I don't know if Factor 5 is telling the truth or helping PS3 sales because they are developing on it.

Define dynamic effects please.
Good for them if they have figured out how to use the SPEs effectively. I would still take it with a grain of salt unless they figured out how to use at least 40% or more of the total SPE (all 6 SPEs) power effectively.

Dynamic effects (such as rain, smoke, snow, clouds, or dust with a changing wind environment at different levels) requires a lot of physics calculations trying to calculate the current position and shape of each individual particle within the changing environment for each new frame. If less than 50% of the cell processor power was used, the same calculations probably could be able to be done with 2 cores of the xenon core.

However, if it did take that much processor power, would it be beneficial for developers using the xenon to produce an application with that much fine dynamic effects at the expense of the actual game play? No one has ever stated that the cell would not excell in certain areas. If a program was based solely on how much dynamic effects could be reasonabably placed on the screen, the cell would probably win every time.

Good for them if they have figured out how to use the SPEs effectively. I would still take it with a grain of salt unless they figured out how to use at least 40% or more of the total SPE (all 6 SPEs) power effectively.

Dynamic effects (such as rain, smoke, snow, clouds, or dust with a changing wind environment at different levels) requires a lot of physics calculations trying to calculate the current position and shape of each individual particle within the changing environment for each new frame. If less than 50% of the cell processor power was used, the same calculations probably could be able to be done with 2 cores of the xenon core.

However, if it did take that much processor power, would it be beneficial for developers using the xenon to produce an application with that much fine dynamic effects at the expense of the actual game play? No one has ever stated that the cell would not excell in certain areas. If a program was based solely on how much dynamic effects could be reasonabably placed on the screen, the cell would probably win every time.

Dynamic Effects though are usually visual in nature though and thus would it be safe to say to be able to produce those visual effects all at the same time, at a higher definition could be a problem if the RSX graphics processor isn't able to produce these effects effeciently?

Also have you seen the Tech Demo for the game Called Alan Wake, coming out on 360 and PC.

Here is a link to it:

http://www.youtube.com/watch?v=XGoiQaMM61E

In this video the producers of the game are using a Quad Core Intel processor.

BTW, I am sure you are aware that the Xenon Cores are able to handle 2 threads at a taime and thus, have 6 threads of infomration going through the three cores.

Dynamic Effects though are usually visual in nature though and thus would it be safe to say to be able to produce those visual effects all at the same time, at a higher definition could be a problem if the RSX graphics processor isn't able to produce these effects effeciently?

Also have you seen the Tech Demo for the game Called Alan Wake, coming out on 360 and PC.

Here is a link to it:

http://www.youtube.com/watch?v=XGoiQaMM61E

In this video the producers of the game are using a Quad Core Intel processor.

BTW, I am sure you are aware that the Xenon Cores are able to handle 2 threads at a taime and thus, have 6 threads of infomration going through the three cores.
I'm not a graphics programmer but I assume that if the CPU rendered any graphics down to a bitmap level (such as what is done by an MPEG2 decoder), the GPU really doesn't have much to do. So I would suspect that Dynamic Effects are can be rendered to the bit map level. In this case, all the GPU needs to do is render the other graphics underneath the dynamic effects. However if there are massive particles defined as mathematical equations, the CPU may have to render them totally to the bitmap level to keep the GPU from bogging down.

Yes, I know about the hardware threads. The developer can also create as many software threads that are desired (only limited by the number that is supported by the operating system). Software threads are not as efficient as hardware threads and are not always executing like hardware threads since the operating system must swap those in/out on a regular basis but once they are swapped in, they become hardware threads. So the primary difference when software threads are added, all threads on that hardware thread become software threads when swapped out but hardware threads when swapped in and the overhead of monitoring and performing the swap is extra overhead.

Here goes a question then, did you watch the Alan Wake video, and what was your assessment of what the programer had to say about what was being done on the Quad Core Processor. How equivlant to a Quad Core Intel Processor without HT is the Xenon with its Multithreading ability, and likewise with Cell?

I know the developer said that this game couldn't be done on a Single Core processor Solution, so I would have to take a guess he means PPE cores since this is what a quad processor would be. If I remember correctly you stated that the SPE's are not equivlant to one PPE Core?


Just trying to get a better grasp of the situation at hand.

It might be assumed that this could mean that basically the SPEs just aren't capapble enough or that game developers just won't be able to code properly to make use of the SPE's fully?

What is your take on this from your academic and technical understanding as one generally don't issue such a statment unless you are trying to cover up some sort of problem?
I don't try to read anything into any statement like that. A slip of the tongue is usually just that. Only time will tell if he was trying to give out a hint of something to come.

I don't try to read anything into any statement like that. A slip of the tongue is usually just that. Only time will tell if he was trying to give out a hint of something to come.

Understood. I might just delete the post actually from this thread.

Understood. I might just delete the post actually from this thread.
It doesn't bother me.

Here goes a question then, did you watch the Alan Wake video, and what was your assessment of what the programer had to say about what was being done on the Quad Core Processor. How equivlant to a Quad Core Intel Processor without HT is the Xenon with its Multithreading ability, and likewise with Cell?

I know the developer said that this game couldn't be done on a Single Core processor Solution, so I would have to take a guess he means PPE cores since this is what a quad processor would be. If I remember correctly you stated that the SPE's are not equivlant to one PPE Core?


Just trying to get a better grasp of the situation at hand.
Hyperthreads or hardware threads main advantage over software threads is that two theads are sharing clocks instead of each taking a large chunk of cpu power at one time. The main issue is that each thread has it's own set of registers, intruction pipeline and controls which control the flow of data to and from the correct register set. So every other clock it used by the other thread to execute or initiates operations. Older processor did not have this capability and could only use software threads where the current register set had to be save to memory and the next register set had to be retrieved from memory and placed in the register set allowing that thread to continue execution. The main advantage is that if one thread stalls waiting on memory or event, the other thread will use its clocks until the other is freed from the stall. Like everything, a hardware thread has its advantages but I wouldn't consider it a big deal.

The Xeon processor is Intels high end workstation and server processor. A quad xeon, I would suspect, would overall blow away the Intel Core Quad, the xenon processor, or the cell processor. Comparing the different processors are dificult since benchmarks for the Intel and PowerPC are usualy geared to make each look the best. Linpack benchmarks (N=1000) is suppose to give a good comparison of how the vector floating point operations perform but manufactuerers have to modify the benchmark to get it working with its vector unit. When the Linpack benchmarks were run on the cell processor, a N=1000 result was not given since I suspect that the benchmark was modified beyond the allowed amount and only numbers were posted for different matrix sizes. It appears that the cell processor averaged between 8-13 GFLOPS per core if the matrix size was 1536x1536 or larger which is pretty good but started to stink as the matrix sizes got smaller. So I don't know what to think about the benchmark.

Since I don't have any hard facts to go by, I can only speculate that an Intel 2.4 GHz Core Quad and the 3.2 GHz cell processor would both be near the top in potiential processing power and the xenon would be furthur down the line. But I could be wrong.

In some operations the SPE core are equivalent to and can sometimes exceed a PPE core. This is primarily in the area of floating point operations. Since the demo was primarily developing AI and graphics, both the cell and the xenon has the potiential of generating those screens (but it isn't really a game). Again the physics code of blowing sands is primarily vector math code. Scenes like that could tax the xenon but with 6 SPE cores and 1 PPE core in the cell, there would probably be enough processor power to generate those scenes. Don't get confused by that since there is much more than just generating dynamic effect graphics in a game application.

Edit: Misread poster and posted about Xeon processor but used the incorrect name. Still refers to the Xeon processor so it makes a little more sense.

Hyperthreads or hardware threads main advantage over software threads is that two theads are sharing clocks instead of each taking a large chunk of cpu power at one time. The main issue is that each thread has it's own set of registers, intruction pipeline and controls which control the flow of data to and from the correct register set. So every other clock it used by the other thread to execute or initiates operations.


That isn't how hardware multi-threading works (that scheme wouldn't work very well). There are different versions. I'm not sure which one the Xenon uses. On a Pentium 4, there is coarse grained multi-threading. When one thread is stalled, waiting on memory, the other thread can switch in. In a long pipe processor like the P4, this can make some sense.

The other extreme is simultaneous multithreading. In this scheme, the processor can actually pick instructions from multiple threads to execute in the execution units of the processor at the same exact time. This type of multithreading provides the most benefit.


The Xenon processor is Intels high end workstation and server processor. A quad xenon, I would suspect, would overall blow away the Intel Core Quad, the xenon processor, or the cell processor.


Xeon is Intel's high end server and workstation class processor, not Xenon.

Xenon is a PowerPC processor that is in the XBox 360.

So the rest of that post doesn't really make sense.

That isn't how hardware multi-threading works (that scheme wouldn't work very well). There are different versions. I'm not sure which one the Xenon uses. On a Pentium 4, there is coarse grained multi-threading. When one thread is stalled, waiting on memory, the other thread can switch in. In a long pipe processor like the P4, this can make some sense.

The other extreme is simultaneous multithreading. In this scheme, the processor can actually pick instructions from multiple threads to execute in the execution units of the processor at the same exact time. This type of multithreading provides the most benefit.

You are correct, I agree there are different forms of hardware threading. I was quoting pretty much how very basic in-order execution processors may work. Any processor can support hardware threading in any way as long as clock sharing ot time sharing is performed and stalls are handled. Since there are so many processors out there, I haven't kept up on how any of them exactly do it. Out-of order execution processors would probably perform much better but all this really depends on the sophistication and abilities of the processor. At a minimum, I suspect that most would like to implement clock sharing (each thread could issue simultanously on the same clock), but with stripped down game processors, who knows how it is really implemented.
Xeon is Intel's high end server and workstation class processor, not Xenon.

Xenon is a PowerPC processor that is in the XBox 360.

So the rest of that post doesn't really make sense.
I responded to too many questions yesterday and I wanted to get that one out of the way before I retired. For some reason since I assmumed that he was talking about the Xeon processor since he was talking also about the Intel Core processor and then I mistakenly used Xenon when I referred to that processor. I responded to too many posts in one day. Thanks, I'm correcting that post even though it still won't make sense to the question asked.

Since Mterzich already explained dynamic effects, I'll just say I have no idea what you guys are talking about. I mean, I know a little about the CPUs and GPUs, but when it comes to some of the effects I've seen in first generation PS3 games as I've explained already here, they look better than what we've seen in first and second generation games on 360, and I would say that's a fact. Is it because you have very skilled developers, like Insomniac coding for PS3 and knows how to do some amazing things with CELL? Probably so:

http://www.spectrum.ieee.org/dec06/4745

To pull this off, they had to learn to squeeze as many calculations per second as they could out of an unfamiliar chip. The Cell is capable of number crunching that would choke other top-of-the-line processors, but only because its architecture is like nothing game developers have ever seen. Each chip contains a single processor that coordinates the work of eight others. And each of those eight differs greatly from the kind of processor you’d find in a PC or game console today. The Cell lets the programmer ask many things of it at once, but it requires those tasks to be molded to the particular abilities of the eight specialized processors. That means programmers will have to learn new ways to code games. It could be that programming the Cell is the most challenging game that developers have ever played, with one big difference: the fate of one of Sony’s most important products hangs in the balance.



Here's a GIF of the snow in Resistance:

http://img.photobucket.com/albums/v500/EvilTaru/snow3.gif


Framerate seems to be a bit fast in that gif, but for the people that have the game and know what I'm talking about, you could just pull the camera back, point at the sky, and stare for a moment being lost and amazed at the same time.

Very good article JackBau3r. I don't know that there were any such game developing companies remaining. There attitude seems reminiscent of companies in the 1980s. They appear to have gotten very good developers that are excited by new things (good programmers can develop at over 10x the speed and quality of an average programmer), have probably worked endless hours, and are hungry to succeed.

Those are the types of companies that Sony and/or Microsoft both need. Normally that type of company is not found with the larger companies.

Have you honesty ever heard of Insomniac? You know, the Ratchet and Clank games on PS2? Yes, they're a very talented developer, and so is Naughty Dog, which is why I was gloating about them earlier in the thread about their new game on PS3. You should see their demo for PS3 Ratchet and Clank video:

http://youtube.com/watch?v=tGH9zBaXBt8


Sorry for the youtube version, but I have to leave soon.

This is an interesting article comparing the XBox 360 and PS3. But what is more interesting are the comments that are posted:
http://blogs.mercurynews.com/aei/2006/12/graphics_resist.html

It talks about the design decisions in Gears of War:
Well, I suppose my surmise is off base here. Tim Sweeney, CEO of Epic Games, came in with an answer on my question about graphics for Gears of War. Here's Tim's answer to my questions:

"The key gameplay elements of Gears of War are design decisions and don't reflect some sort of hard-coded design parameters in the underlying technology. We're also using Unreal Engine 3 as the basis for Unreal Tournament 2007, which has very fast player movement speeds and a very different system for aiming and damage.

With Gears, we aimed to create a realistic feeling of fighting in a war. This meant realistic player movement speeds, realistic damage (thus necessitating the cover tactics), less accurate shooting when moving, etc. This was a huge change from Unreal Tournament, where the player runs around at 30 miles per hour and can take multiple rocket hits without dying.

The motion blur effect we apply while roadie-running is actually more expensive than rendering the image unblurred. We do it to give the player sense that he's moving at a running pace with impaired aim."

Have you honesty ever heard of Insomniac? You know, the Ratchet and Clank games on PS2? Yes, they're a very talented developer, and so is Naughty Dog, which is why I was gloating about them earlier in the thread about their new game on PS3. You should see their demo for PS3 Ratchet and Clank video:

http://youtube.com/watch?v=tGH9zBaXBt8


Sorry for the youtube version, but I have to leave soon.
Several times during the thread I stated I don't own any console and I'm not and have never been a gamer. So no I have not heard about any of those. I'm only looking at the consoles from a technical point of view with the emphasis how developers can develop on those consoles.

Well that's odd, lol.

I don't see it has odd. I see mterzich's view pretty much impartial since he doesn't own either. That's a good thing.

It's definately impartial, because since he doesn't know anything on future games for 360 or PS3. Sure, you can look at specs on some blogspot that says "x" is better than "y", but in the end, it'll come down to the games.

It's definately impartial, because since he doesn't know anything on future games for 360 or PS3. Sure, you can look at specs on some blogspot that says "x" is better than "y", but in the end, it'll come down to the games.

True. But "coming down to the games" is related to the power, design, and programmability of the console. How good the games are will be related to this.

lol. But if he doesn't know what type of games will be released in the future on each console, how will he know if the specs in his OP will be applied to those future games? It pretty much cancels out.

well this is a processor comparison, not a game comparison. At least that's what the post says

lol. But if he doesn't know what type of games will be released in the future on each console, how will he know if the specs in his OP will be applied to those future games? It pretty much cancels out.
I don't believe anybody knows for sure what type of games will be successful in the future. Snowflakes and dust storms are nice but that alone will not sell the games.

My main concern with the cell processor was not that the cell processor used SPEs but that it appeared it was very heavily weighted in favor of SPEs. In my opinion, the PS3 would probably have an advantage over the 360 if it had 2 PPE cores and 6 SPE cores (instead of 1 PPE and 8 SPE cores).

Currently it appears, most developers load routines into cell processors that remain resident in the SPEs throughout the game. This creates for difficulty in effectively utilizing the full potiental processor power of the SPEs. Since the total memory of the 6 SPEs are only about 1.5 MB, developers attempt to find enough code which can reside with its data in 256 KB of memory out of the 100 MBs or more of code in the game application that will heavily utilize the SPEs processor power. SPEs do not have hardware threads and software threads would probably be unlikely since each cell would probably require a mini operating system to support software threads. So the SPE is currently single threaded and the PPE will have to wait until the SPE is finished with the current routine before it can notify the SPE to start executing another routine.

In contrast, as many threads as desired can be assigned to each core of the xenon processor. Since all game code and data remains resident in main memory, the controlling PPE can at any time (as long as the data to be processed is available) start up a new thread and execute a specific routine using the data in a certain part of memory and save the results in another part of memory. The controlling PPE can possibly even startup another thread executing the same routine but with a different set of data inputs and results. Possibly 50 or more threads could be executing at one time just saturating all the 3 PPE cores.

To possibly saturate the SPE cores, some sort of paging system would appear to be needed. However, if code and data is paged in and out of the PPE memory, a point will occur were there will be diminishing returns since the main bus may get bogged down, overall throughput may decrease, the PPE may use too much processor power to perform the paging operations, and program latency may become unacceptable (the SPE cannot startup until the code and data is paged in).

Both processors have both their possitives and negatives. The xenon processor does not have the shear processor power of the cell processor but it is easy to program and saturating the 3 available cores should be fairly easy. The cell processor has a large amount of processor power but it is difficult to fully tap as well as difficult to fully saturate the cores.

Both of the processors are good processors. If Sony can convince the developers to spend the extra required time and the developers figure out ways to heavily utilize the SPEs to perform operations that are desired by the developer (important that it is the desired operations), Sony should do very well. Microsoft has the advantage of simplicity so development costs are reduced and everything was know about how to develop code for the 360 as soon as the console was released. Due to that fact plus being to market early, it would appear that Microsoft has the current advantage. Micrsooft also appears to be able to more easily line up exclusive developers. In the future, exclusive game development (exclusive killer games) will probably make the difference.

Both of the processors are good processors. If Sony can convince the developers to spend the extra required time and the developers figure out ways to heavily utilize the SPEs to perform operations that are desired by the developer (important that it is the desired operations), Sony should do very well. Microsoft has the advantage of simplicity so development costs are reduced and everything was know about how to develop code for the 360 as soon as the console was released. Due to that fact plus being to market early, it would appear that Microsoft has the current advantage. Micrsooft also appears to be able to more easily line up exclusive developers. In the future, exclusive game development (exclusive killer games) will probably make the difference.

You have hit the nail on the head though with this assessment. It is a matter of whether or not game developers are willing to take the risk to put forth the effort, and the money, to take advantage of what ever the Cell processor is capable off.

Game costs are rising where to even make a high quality game, Sony apparantly is demanding more quality out of there developers, the reported costs is roughly 20 million dollars. Developers have stated that to recooperate costs that of just developing a game 500k copies of a game will have to be sold.

Game developers are apparantly already beginning to rethink about developing games or keeping games exclusive to the PS3 due to these issues of time, cost, and lack of techical/software support for developing on the Cell.

Just wanted to bring you up to speed on some of the current happenings that are taking place in the gaming world since you don't pay attention to it.

The way I see it though development costs will be to high and the rewards to low in the end for developers to tap the PS3 Cell as to do so will just require to many man hours for quite possibly unknown results, not to mention those bottleneck issues and the like you brought up and the various workarounds that need to be figured out to utilize the processor.

I don't believe anybody knows for sure what type of games will be successful in the future. Snowflakes and dust storms are nice but that alone will not sell the games.


You're trying to compare processors here. Now it's about which game will sell? Make up your mind. You can type up 500 words speculating which is more powerful, but fact is, CELL is more powerful than Xenon, and the games prove it, i.e. creating realistic snow, weather effects, animation and physics. Highly acclaimed developers like Naughty Dog, Insomniac, Konami, and SquareEnix have created their own next gen. engines, and it looks like it'll be proven that their engines are technically better than what most developers are doing with the UE3.

You can make fun snowflakes and duststorms, but in the end, it adds another level of immersion and realism. I don't think nobody wants a next generation game with unrealistic post processing effects. Anyway, here's another example of how talented Insomniac is:

http://members.shaw.ca/VTXtreme/PS8.jpg

It's hard to get a better picture of the how realistic the glass looks when you at windows in Resistance, but the detail is amazing.

I don' t know, there are PC games and simulation have the weather elements: rain, , wind, snow, 3D clouds. One flight simulation downloads weather in real-time and updates it.

There's more to a game than just the CPU IMHO. Sure, if you were back in the 90s, it was all about the processing power. Since the inception of the 3D cards from 3DFx and Rendition, times have changed in which it shows that you need GPU performance as well.

The problem with the PS3 is that it seems Sony waited too long with the GPU selection or design. They decided on an 'older' GPU from Nvidia which was based on DX9 and has a major issue if HDR and anisotropic filtering are both enabled. The result is a major performance drop, so you can only use one. Now the newer Nvidia GPUs don't have this problem. Perhaps when the GPU begins to slow the CELL will give some boost back????


Now with the 360, you have a GPU that was designed for DX10. DX10 gives you all the benefits of implementing current crop of features (FSAA, AF, bumpmap, etc.) with out taking a major FPS hit like the older DX9 GPUs experienced. Currently, the 360 is using a DX9 hybrid and there is a planned DX10 update coming.

Shots between DX9 and DX10 GPUs using the Crytek 2 Engine

DX9 is always on the left

EDIT: Check the CRYSIS website forum for DX9 vs DX10 images.

Technical features of DX10

Heres some of the technical features:

-Dense, lush vegetation made possible by SM2.0b geometry instancing
-HDR lighting with realistic light flares
-Realistic environments with proper lighting, reflections and refractions
-Water with reflections and refractions
-Shadows
-Soft shadows use high-precision shadow maps
-Characters self-shadow for ultra-realistic lighting
-Full-precision per-pixel lighting
-Characters
-Highly detailed bump-mapped and normal-mapped characters
-Realistic facial animations
-3Dc helps to reduce memory requirements and raise rendering quality for highly detailed characters
-Post-processing numerous advanced visual effects
-HDR light glows
-Predator effect with cool plasma-like glow when monster gets neutralized
-Special night-vision mode for security guys
-Time-space warp mode
-Depth of field effect used throughout the demo to draw the attention to particular parts of the scene make it a true cinematographic experience
-Motion blur is another cinematographic effect to emphasize the speed of the rocket at the end of the demo
-Large variety of advanced shaders to create realistic look for different material and surface types
-Complex materials like water using well over 50 pixel shader instructions
-Long vertex shaders and high density geometry going beyond the capabilities of previous generations of graphics hardware
-Almost 300 different pixel shaders and over 300 distinct vertex shaders
-Over 200MB of graphics data per frame
-Full support for High Def resolutions for maximum quality
-Use of high quality anti-aliasing to minimize geometry aliasing

mzterich

since you seem to be curious about CELL, I suggest going over to the console tech forum over at beyond3d.com. lots of excellent technical discussion on the capabilities of CELL, as well as real game devs posting from time to time. I think your opinion of CELL will be different after spending a bit of time there.

in fact, I'd encourage anyone to take a look over there, but it's very technical.

Shots between DX9 and DX10 GPUs using the Crytek 2 Engine

DX9 is always on the left
Technical features of DX10

Heres some of the technical features:

-Dense, lush vegetation made possible by SM2.0b geometry instancing
-HDR lighting with realistic light flares
-Realistic environments with proper lighting, reflections and refractions
-Water with reflections and refractions
-Shadows
-Soft shadows use high-precision shadow maps
-Characters self-shadow for ultra-realistic lighting
-Full-precision per-pixel lighting
-Characters
-Highly detailed bump-mapped and normal-mapped characters
-Realistic facial animations
-3Dc helps to reduce memory requirements and raise rendering quality for highly detailed characters
-Post-processing numerous advanced visual effects
-HDR light glows
-Predator effect with cool plasma-like glow when monster gets neutralized
-Special night-vision mode for security guys
-Time-space warp mode
-Depth of field effect used throughout the demo to draw the attention to particular parts of the scene make it a true cinematographic experience
-Motion blur is another cinematographic effect to emphasize the speed of the rocket at the end of the demo
-Large variety of advanced shaders to create realistic look for different material and surface types
-Complex materials like water using well over 50 pixel shader instructions
-Long vertex shaders and high density geometry going beyond the capabilities of previous generations of graphics hardware
-Almost 300 different pixel shaders and over 300 distinct vertex shaders
-Over 200MB of graphics data per frame
-Full support for High Def resolutions for maximum quality
-Use of high quality anti-aliasing to minimize geometry aliasing

I can't wait for Crysis to hit. Also the point your bring up about the graphics processor is indeed I think the more critical issue here then processing power anyway. I mean it doesn't matter what those various effects the cell can process if the textures are of a lower quality due to the RSX lack of being based on the future architecture for PC Cards.

Also I am taking a guess that Jack Bauer the user in this thread hasn't played the opening sequence of Act 3 on Gears of War were there is a Thunderstorm taking place. Or when the weather changes during Oblivion. I think Alan Wake is really going to show of some amazing real time effects as well.

Also it should be stated to avoid confusion that Xenos is capable of many of the DX 10 feature set, but not all.

To add to that, the over all architecure of the system board and how all the components work together to esnure optimal performance of all components shouldn't be overlooked.

To bring up an analogy:

What does it matter if you have a 500 HP engine in the car, if it is filled with sludge.

You're trying to compare processors here. Now it's about which game will sell? Make up your mind. You can type up 500 words speculating which is more powerful, but fact is, CELL is more powerful than Xenon, and the games prove it, i.e. creating realistic snow, weather effects, animation and physics. Highly acclaimed developers like Naughty Dog, Insomniac, Konami, and SquareEnix have created their own next gen. engines, and it looks like it'll be proven that their engines are technically better than what most developers are doing with the UE3.

You can make fun snowflakes and duststorms, but in the end, it adds another level of immersion and realism. I don't think nobody wants a next generation game with unrealistic post processing effects. Anyway, here's another example of how talented Insomniac is:

http://members.shaw.ca/VTXtreme/PS8.jpg

It's hard to get a better picture of the how realistic the glass looks when you at windows in Resistance, but the detail is amazing.
First of all, I wasn't making fun of a processor to produce Dynmanic Effects in mass. Secondly you were the one who brought up the issue of games and then critized me when I responded. Up until you last post, you were using a reasonable degree of discussing the pros and cons. This thread if at all possible, unlike so many others, will not deteriorate into fanboy attacks. There are plenty of other threads for that.

http://members.shaw.ca/VTXtreme/PS8.jpg

Wow, that brick wall texture on the right side of the window looks horrible.

http://members.shaw.ca/VTXtreme/PS8.jpg

Wow, that brick wall texture on the right side of the window looks horrible.

Well, I say lets get back on topic and do our best to avoid getting into an all out flame war and console war scenario.

Have you gained any further insights or understandings today mterzich?

Have you looked into any of the developer tools that people are you using?

I know Microsoft has XNA out a developer tool that anybody is actually able to download and all it requires is C# Express.

I have it downloaded but, since I need to learn the language, haven't done much with it yet.

I know you stated that your aren't interested in games, yet I am not sure if looking at what can be done with program could halp you with your analysis at all?

Nah, thats rubbish mate. If I was a console developer and knew anything about the in's and outs of this hardware I'd spill the beans, even if it meant anonymously. Let's not get crazy - books will be published for these "computers", textbooks, whitepapers. It's not a big secret when it's in plain site.

Muz
256K is not that much memory it is about 16000 vertices, assuming double precission. That ain't squat in modern day video games, that is probably about 1 character of data, and that is only vertexs no textures. 256K is nothing especially with today's games requiring large dataset's just to be able to represent one frame, average 2k to 200k of vertices, alone. DMA transfering to and from a buffer the PPE can only access will kill the architecture. The SPE's are not designed for this type of application. It's all marketing hype. Yeah if this were the days of dos then these SPE's would be powerhouses. Untill the SPE's can access more than 2mb of ram they will be a useless piece in this iteration of the console wars. Just don't see any use for them, maybe good for sound, or somthing but graphics no. You cannot use them for geometry setup, poor branching eliminates them from AI use, and having to swap in and out data via a dma channel will require so much overhead maintaining coherency, and execution order.

I cannot belive that the SPE's don't have any paths to some volume of memory, this in effect makes them almost useless. I don't know maybe I'm missing somthing but I have writen several 3d engines from the ground up, and know how they work just cannot see what the SPE's bring to the table.

http://members.shaw.ca/VTXtreme/PS8.jpg

Wow, that brick wall texture on the right side of the window looks horrible.


I think that pic shows a very nice shattered glass effect... and a lot of ugliness through the rest of it.

I'd ignore it though, someone is trolling and refuses to believe that those pretty effects aren't capable of being done on both systems. Good developers find a way, and MS made things easy and did a good job on memory and bandwidth (which is huge). Both systems are going to be good enough.

Forgot to mention, thanks to mterzich for very good analysis and very unbiased.

http://members.shaw.ca/VTXtreme/PS8.jpg

Wow, that brick wall texture on the right side of the window looks horrible.


Yeah. The textures really are a hit or miss in the game:

http://smallvillestories.free.fr/images/Afrika/restank6.jpg
http://smallvillestories.free.fr/images/Afrika/resistance2parquet.jpg

Yeah. The textures really are a hit or miss in the game:You're contributing precisely zero to this thread and are clearly outclassed. It's time to take your trolling back to the PS3 forums.

512 MB main memory that is shared by all three cores.


I heard a recent interview with Cliffy B (Gears of War developer). One of the interesting comments was that he stated that MS doubled the memory of the 360 when his team showed them what Gears (an early prototype) could look like with 512 vs. the original 360 spec of 256 MB of main memory. Increasing the memory cost MS some $$s, but he stated that this was really worth it. SJ

Yeah. The textures really are a hit or miss in the game:

http://smallvillestories.free.fr/images/Afrika/restank6.jpg
http://smallvillestories.free.fr/images/Afrika/resistance2parquet.jpg

OT: I agree that Resistance has some great textures especially in the later levels. I noticed the glass too when I was playing it. I zoomed in on it and I was like, "I've never seen glass like that before in a video game" Lots of people probably judge the game on it's earlier levels which consist of washed out greys and reds that give it a dull look.

You're contributing precisely zero to this thread and are clearly outclassed. It's time to take your trolling back to the PS3 forums.


Yeah. I've clearly contributed nothing to this thread, yet mterzich liked the article about CELL that I posted earlier. I've already claimed I don't know much about CELL and Xenon, but I know about some of the PPE's both are capable of in game, like some of the examples I've posted. If there's someone who hasn't contributed nothing to this thread, it's you. Nice try.

mterzich. If you want good discussions on CELL benchmarks or Xenon's, then go here:

http://www.beyond3d.com/forum/showthread.php?t=36335

http://rian.s26.xrea.com/nicky.cgi?DT=20061121A#20061121A

・Dhrystone v2.1
PS3 Cell 3.2GHz: 1879.630
PowerPC G4 1.25GHz: 2202.600
PentiumIII 866MHz: 1124.311
Pentium4 2.0AGHz: 1694.717
Pentium4 3.2GHz: 3258.068

・Linpack 100x100 Benchmark In C/C++ (Rolled Double Precision)
PS3 Cell 3.2GHz: 315.71
PentiumIII 866MHz: 313.05
Pentium4 2.0AGHz: 683.91
Pentium4 3.2GHz: 770.66
Athlon64 X2 4400+ (2.2GHz): 781.58

・Linpack 100x100 Benchmark In C/C++ (Rolled Single Precision)
PS3 Cell 3.2GHz: 312.64
PentiumIII 866MHz: 198.7
Pentium4 2.0AGHz: 82.57
Pentium4 3.2GHz: 276.14
Athlon64 X2 4400+ (2.2GHz): 538.05




Here's a discussion on Xenon being just a modified G5 triple core:

http://www.beyond3d.com/forum/showthread.php?t=31125&highlight=Xenon

G5 can manage 5 instructions per cycle and has the next units:

-2 Branch Units
-2 Load/Store Units
-2 Integer Units
-2 FPU
-2 VMX

Xenon can manage 2 instructions per cycle and has the next units:

-1 Branch Unit
-1 Load/Store Unit
-1 Integer Unit
-1 FPU
-1 VMX-128

The FPU+VMX-128 part has been revamped but I don´t see how is possible that a new CPU with half units could have more transistors than a more complex CPU x3 with the same Shared L2 Cache.

G5 Logic without L2 Cache= 34 milions of transistors.
G5 Triple Core Logic= 102 milions of transistors
G5 Triple Core with Shared L1 Cache= 151 milion transistors
What Xenon documents say= 165 milion transistors.

Obviously that a Xenon core is less powerful and obviously it needs less transistors but the documentation says that the transistors are more than a possible G5 triple core with Shared L2 Cache.

I am sure that we have 3 G5 inside the Xenon core and thanks to the disables OOE the system can only do 2 instructions per clock and this is when having 2 Integer Units, 2 FPU, 2 VMX... Become useless. And this tells how Microsoft only needed 14 months for finishing the project.

PD:It can be applied to the PPE.


Also I am taking a guess that Jack Bauer the user in this thread hasn't played the opening sequence of Act 3 on Gears of War were there is a Thunderstorm taking place. Or when the weather changes during Oblivion. I think Alan Wake is really going to show of some amazing real time effects as well.


Yes, I have. You really get sucked in with such detail that it's just so surreal.

What I'm really concerned about next generation is animation. We've seen impressive animation last generation, like in ICO, SotC, or Wind Waker. Those games moved so fluidly. I really do hope it gets taken to the next level this generation. My question is, which CPU will have the advantage when it comes to animation? I'm seeing some incredible things being done with Ubisoft's schimiter engine(Assassin's Creed), and it's rumored that Splinter Cell 5 will have AMAZING animation where you won't be able to find Sam Fisher having the same pose when jumping. Then there's Naughty Dog's new game sporting the ICE engine. I dunno, It seems like it'll just come down to the developers. Just look at Team ICO. Developed on an inferior console out of the three last generation, and created some surreal games.

256K is not that much memory it is about 16000 vertices, assuming double precission. That ain't squat in modern day video games, that is probably about 1 character of data, and that is only vertexs no textures. 256K is nothing especially with today's games requiring large dataset's just to be able to represent one frame, average 2k to 200k of vertices, alone. DMA transfering to and from a buffer the PPE can only access will kill the architecture. The SPE's are not designed for this type of application. It's all marketing hype. Yeah if this were the days of dos then these SPE's would be powerhouses. Untill the SPE's can access more than 2mb of ram they will be a useless piece in this iteration of the console wars. Just don't see any use for them, maybe good for sound, or somthing but graphics no. You cannot use them for geometry setup, poor branching eliminates them from AI use, and having to swap in and out data via a dma channel will require so much overhead maintaining coherency, and execution order.

I cannot belive that the SPE's don't have any paths to some volume of memory, this in effect makes them almost useless. I don't know maybe I'm missing somthing but I have writen several 3d engines from the ground up, and know how they work just cannot see what the SPE's bring to the table.
Nice to have someone on the thread who has had experience developing graphics.

Without some mechanism to access main memory directly, you are correct, the SPE is limited. When it was stated that the PS3 used three SPEs to decode MPEG4 streams, I couldn't figure how they accessed the 6MB (3MB for 720p) video buffer in main memory since it couldn't be transferred to the SPE. Wouldn't it kill the performance writting (using DMA) about 8 blocks of video data for every macroblock? And if motion compensation is being performed, reading (using DMA) each block is done before the writting is performed. When the developers wrote the decoder, the performance was terrible as expected. Eventually, they found an acceptable solution for this problem (wouldn't work for graphics) by redefining the raster scan video buffer so that data was more continious so microblocks would require less DMA reads and writes. After the decoding for each frame was completed, the PPE would then have to copy and convert that video buffer to a real raster scan video buffer that could be sent to the GPU. For MPEG decoding, this concept along with making sure everything was aligned correctly gave acceptable performance but using a technique like that in video games would seem to appear to cause much to much overhead (DMA and copying and converting the video buffer).

Although increasing the SPE memory size would probably help quite a bit, until there is some mechasim to have high speed access to main memory (possibly a common L2 cache), there will still be limitations on the SPE.

JackBau3r,

The xenon processor being stripped down with in-order execution (lack of parallel execution units) and reduced branch prediction capabilities was discussed extensively in the original posting. This gives the xenon processor only about 40% of the processor power of a Apple G5 at the same clock rate. The same is true about the PPE core in the cell processor. The main difference between the PPE core in the cell processor and the xenon processor is that the xenon processor has an enhanced register file for each core to improve vector performance and the GPU can access the L2 cache.

This is why some developers initially stated that the xenon processor didn't perform very well since 1 PPE core only had about 2x the processor power of the original xbox.

When I originally made this thread, I also wondered why the xenon processor used so many transistors (165 million) since it stripped out so much circuitry. I was expecting it to have about 100 million transistors. I could possibly see it having that many transistors if it had 2 MB L2 cache but that is not the case. We know that IBM has not yet been able to get a standard PowerPC yet working at 3.2 GHz so maybe they had to strip out that ciruitry to get it working at the frequency. Although it seems it would have been much better if the 360 had a standard 2.5 GHz PowerPC processor, I don't know all the ins and outs concerning cost and other factors in deciding on which processor to use.

Even if Microsoft would have used a standard 3-code 1.6 Ghz PowerPC, the performance would probably have been better but from a marketing point of view they may have been clobbered since many perceive the clock rate as being the most important.

Yeah. I've clearly contributed nothing to this thread, yet mterzich liked the article about CELL that I posted earlier.It was certainly an effective way to get him to believe you were serious about the discussion before you started in with the fanboism. Now that you've been called on it, at least you're getting back on track, which I applaud.

256K is not that much memory it is about 16000 vertices, assuming double precission. That ain't squat in modern day video games, that is probably about 1 character of data, and that is only vertexs no textures. 256K is nothing especially with today's games requiring large dataset's just to be able to represent one frame, average 2k to 200k of vertices, alone. DMA transfering to and from a buffer the PPE can only access will kill the architecture. The SPE's are not designed for this type of application. It's all marketing hype. Yeah if this were the days of dos then these SPE's would be powerhouses. Untill the SPE's can access more than 2mb of ram they will be a useless piece in this iteration of the console wars. Just don't see any use for them, maybe good for sound, or somthing but graphics no. You cannot use them for geometry setup, poor branching eliminates them from AI use, and having to swap in and out data via a dma channel will require so much overhead maintaining coherency, and execution order.

I cannot belive that the SPE's don't have any paths to some volume of memory, this in effect makes them almost useless. I don't know maybe I'm missing somthing but I have writen several 3d engines from the ground up, and know how they work just cannot see what the SPE's bring to the table.

What do you think of the two GPUs found in the systems?

Also what about the overall architecture of the system boards?

Some of you may be thinking that Sony and IBM didn't understand what they were doing when they designed the cell processor. That is not true. IBM wanted a processor of that design for its blades in its massively parallel systems (inexpensive design for that type of processing) and Sony wanted a design with a large amount of processor cores.

But couldn't Sony have requested a design with more local store memory? It may have been possible to increase the 256 KB of SRAM to 512 KB but what would have been the benefit since the memory size would have still very limited. It probably wouldn't have made that much difference on a cost/benefit basis having 512 KB of memory instead of 256 KB.

What if the design allowed the SPEs to access main memory directly. Is that possible and wouldn't that solve most of the issues with the design? Yes, if direct access to main memory was allowed for data, even the limited local store memory size would become much less important. Implementing direct acces to main memory is actually quite a simple change but unfortunately without major changes, the performance would be extemely poor. Main memory has a very long latency each time it is referenced (100-300 clocks) and is usually referenced in blocks of data (128 bytes, 256 bytes, 512 bytes, 1024 bytes, etc) to spread the latency time over the full block. Processor cores do not reference data in blocks but reference data a byte, word, double word, or quad word at a time. And everytime it references memory, it will have the latency. So therefore, the performance of the processor could decrease by as much as 500x referencing the data in such small memory chunks.

So what would have to be implemented to reduce the effects of latency on the processor? Well a common L2 cache for all 8 SPEs of at least 2 MB would have to be implemented. This hides most of the long latency of the main memory but it still produces a latency of 5-15 clocks so now we would have the cores executing at about 5-10x slower than would be acceptable. To eliminate that latency, at least 16 KB of L1 data cache would have to implemented for each SPE. If all that was done, the SPEs would have very fast access to main memory. This hides the latency since cache accesses memory in blocks in anticipation that individual elements of those blocks will be wanted in the future.

Cache memory takes up a large amount of real estate on the die and that much cache would probably increase the amount of transistors for the cell by over 100 million transistors. It probably wouldn't have been possible to put that many more transistors on that die making it impossible.

Couldn't they put the L2 cache external to the processor chip? About 15 years ago that was done but things have become more complicated since then. The L2 cache has to be able to commnicate with other devices (PPE L2 cache and DMA devices) and if it doesn't, memory can end up being reflected differently in each of the devices (cache coherency). It still may be possible but I'm not sure that the speeds needed for these cross checks can be done with an external device. I suppose a simplistic approach could be taken in that the external cache as well as the PPE internal L2 cache could be forced to write all dirty data to main memory whenever a DMA operation started or a SPE started and stopped to maintain cache coherency but this could possibly add latency to the DMA device, SPE, and/or PPE core.

So it wasn't that Sony wanted exactly this processor but had to settle for it when all other options were exhausted to create a 9 core processor. Instead Sony settled for the cheap way of accessing main memory by using DMA. DMA accesses memory in blocks reducing the latency per byte but usually causes the SPE to wait until the complete block has been transferred. Also DMA is not automatic and has to be programmed each time to transfer a block of memory.

Xbox 360 vs. PS3 graphics comparison (http://www.gamespot.com/features/6162742/p-8.html) on Gamespot is a reminder of the importance of developers having the time and skill to maximize each architecture:

Conclusion
The Xbox 360 had better graphics in almost all the games we examined. The 360's biggest victories were in Madden 07 and Fight Night Round 3, where the differences in texture detail and lighting stood out in our comparison shots. We couldn't capture this in the screenshots, but the Xbox 360 games generally offered better framerates too.

We're going to give the PS3 the benefit of the doubt in this initial round since developers might need more time to figure out how to maximize performance from the Cell and the RSX.

Xbox 360 vs. PS3 graphics comparison (http://www.gamespot.com/features/6162742/p-8.html) on Gamespot is a reminder of the importance of developers having the time and skill to maximize each architecture:


Thanks, but I have already seen this, if I wanted a simple general answers and links to gamesites were they have side by side shots that really prove nothing i wouldn't have bothered asking the question.

Since XYZVector seems to be more of graphics programmer and might have some more insight into this end of the discussion, about the GPU's of each system. It would be interesting to see his point of view on matters from the architecture of the system board, the architecture of each chip design, what the Unified Shader Architecuture of the Xenos brings to the table, and how each GPU corresponds to the CPU?

mterzich has already posted some of his insights on the matter, but also has stated that he is more of a regular programmer and isn't as knowledgeable on this issue, I hope my assessment of that is correct mterzich. So it would be a great insight if XYZVector could possibly add some more expertise insight into the matter.

Anyway I think I might copy and paste all of your posts mterzich into Word and give each one a Title, this way I can go back to it and refer to points you might have already made and just reread the posts in more detail.

Xbox 360 vs. PS3 graphics comparison (http://www.gamespot.com/features/6162742/p-8.html) on Gamespot is a reminder of the importance of developers having the time and skill to maximize each architecture:
NFS had better textures on the PS3 version, but better lighting on the 360. In FNR3 the fighters on the PS3 look better, but the lighting and background look better on the 360. I typically see most games are a trade off in some way. Also of note in the hanger arena for FN3 the light shinning through the doors on the 360 is annoyingly bright and makes me wonder if it was toned down on purpose in the PS3 version.

NFS had better textures on the PS3 version, but better lighting on the 360. In FN3 the fighters on the PS3 look better, but the lighting and background look better on the 360. I typically see that most games are a trade off in some way. Also of note in the hanger arean for FN3 the lighting shinning through the doors on the 360 is annoying bright and makes me wonder if it was toned down on purpose in te PS3 version.


This is what I'm confused about. How is the 360 version of FNR3 better on 360, when the PS3 version has better shaders and the lighting is more "natural" looking, than the bloom lighting on the 360 version? Even EA said the PS3 version looks better than the 360 due to having more time to develop the game.


360 version:

http://xbox360media.ign.com/xbox360/image/article/682/682250/ea-sports-fight-night-round-3-20060119011912045.jpg

PS3 version:

http://ps3media.ign.com/ps3/image/article/739/739189/ea-sports-fight-night-round-3-20061016111558441.jpg
As for games multiplatform games looking better on 360, I wouldn't doubt it. Both consoles are different to program for, with 360 being the easiest.

This thread is fantastic. Good job mterzich for doing all the research and analyzing of the datat that you have found out. As somebody suggested you should seriously think about getting that information into a more Coherent form on a Website or Blog.

I am very tempted to even to it myself.
I probably won't put the information on a web site or blog but if you or anyone else desires to extract the information and post it in other places, you are very welcome to do so.

It appears that some of the information is already getting distributed on the web. I accidently ran accross a US and German web site that had the original posting when I was doing a search for some information. Today I did a search and found about 10 web sites in different countries that had the original post. The following is a link to the google search.

http://www.google.com/search?sourceid=navclient&ie=UTF-8&rls=GGLG,GGLG:2005-44,GGLG:en&q=%22One+general+purpose+PPE+core+that+is+used+for+the+OS+an d+the+game+application%22

Wow! Talk about spreading around. :)

This is what I'm confused about. How is the 360 version of FNR3 better on 360, when the PS3 version has better shaders and the lighting is more "natural" looking, than the bloom lighting on the 360 version? Even EA said the PS3 version looks better than the 360 due to having more time to develop the game.


360 version:

[IMG]http://xbox360media.ign.com/xbox360/image/article/682/682250/ea-sports-fight-night-round-3-20060119011912045.jpg[/IG]

PS3 version:

[IMG]http://ps3media.ign.com/ps3/image/article/739/739189/ea-sports-fight-night-round-3-20061016111558441.jpg[/IG]
As for games multiplatform games looking better on 360, I wouldn't doubt it. Both consoles are different to program for, with 360 being the easiest.
Yeah, I preferred the lighting on the PS3 version as well. I also believe you're right about multiplatform games looking better on the 360. They will most likely get released before their PS3 counterparts as well due to the 360 being easier to program for.

I probably won't put the information on a web site or blog but if you or anyone else desires to extract the information and post it in other places, you are very welcome to do so.

It appears that some of the information is already getting distributed on the web. I accidently ran accross a US and German web site that had the original posting when I was doing a search for some information. Today I did a search and found about 10 web sites in different countries that had the original post. The following is a link to the google search.

http://www.google.com/search?sourceid=navclient&ie=UTF-8&rls=GGLG,GGLG:2005-44,GGLG:en&q=%22One+general+purpose+PPE+core+that+is+used+for+the+OS+an d+the+game+application%22

No worries, this would be for my own personal use and I don't intend to blog it or website archive it as of now and wouldn't do so without your strict permission.

Thanks, but I have already seen this, if I wanted a simple general answers and links to gamesites were they have side by side shots that really prove nothing i wouldn't have bothered asking the question.



Understood that you are looking for a more "beyond3d" type technical discussion, but I posted to the general topic of the thread moreso than directly to any question of yours.

At the end of the day any technical differences in each system, processor or otherwise, will only matter to the degree that game developers are able to take advantage of them and produce some results that consumers (such as gamespot reviewers) actually see in the games.

This is what I'm confused about. How is the 360 version of FNR3 better on 360, when the PS3 version has better shaders and the lighting is more "natural" looking, than the bloom lighting on the 360 version? Even EA said the PS3 version looks better than the 360 due to having more time to develop the game.

I've heard some reviewers/commentators comment that mainly what EA did for FNR3 is focus all the detail on the boxers. That the crowd is now blurry cardboard cutouts but the boxers look better and the new 1st person mode adds a new gameplay factor (the blurring effects as you take damage). But yeah most people say it is an obvious improvement, even if the main changes are LoD and camera angles.

Some like to point to the fact that in the cell processor SPEs use fast SRAM as the local store memory and the xenon processor uses slow DRAM type memory as its main memory and therefore this gives the SPE a tremendous performance advantage. Although it is true that SRAM will perform much faster than DRAM type of memory, with proper implementation of instruction prefetch, branch prediction, instruction pipeline, and cache, the difference in performance is usually slight.

Sequential Instruction Exection

When instructions are executing sequentially, the instruction prefetch keeps the instruction pipeline full of instructions that may possibly be used in the future.
Seldom does a stall occur when executing sequentally.

Branch Instruction Execution

With a good implementation of branch prediction, normally a stall will not occur.

Writting data to memory

Usually writting data to memory does not have any affect on the performance of the processor.
Data only needs get to the L1 cache and the L1 cache will handle getting the data to the L2 cache and eventaully to main memory.

Reading data from memory

This is where stalls can occur most frequently.
With proper data block look-ahead in the L1 cache, stalls of this type can be drastically reduced.


Usually if all of the above is implemented properly, the difference betwen DRAM access with proper cache and SRAM access is only a few percent difference in the overall performance of the processor.

Generally accepted thinking is that 512 KB L2 cache and 32 KB of L1 instruction and data cache each is sufficeint per core to have very good performance as related to memory access. Any amounts above those are generally reserved for very high performance processors continuiously executing many applications simulatnously.

In the area of L2 cache, the xenon processor appears to be slightly short with only 333 KB of L2 cache per core. If the xenon processor was used as a desktop processor, this amount may be slightly too small but since the xenon processor is used as a game console, only one application is executing at a time so the amount is probably sufficient to maintain good performance.

The whole concept of cache is that data and instructions will normally be available in the L1 cache when needed by the processor. If the data is not in the L1 cache, it expected that it will be in the slightly slower L2 cache, and only infrequently will a longer stall occur waiting for the data to be retrieved from main memory. Due to the way developers develop code and the way compilers work, the concept usually works very well making it usually appear that memory is immediately available whenever it is needed.

In the cell processor, data has to be moved to and from the SPE memory causing a large performance hit. In the xenon processor, moves of data blocks are seldom done because of the performance hit that would occur but instead only a pointer to the data is passed to another core/thread to process the data.

Although the a SPE has guaranteed overall faster acess to its SRAM memory, a xenon processor core exceeds the overall memory performance of a SPE and doesn't have to worry about the memory limitations of its memory.

http://img70.imageshack.us/img70/500/fightnightcompairbp3.jpg

I think this person did a fairly good job showing the differences.

Also, it should be stated that EA did have an extra year to work on the PS3 Version.

It would be interesting if FNR3 360 had been delayed to release with the PS3 version if the 360 version could have the aspects that make the 360 version stand out in addition to those few details that make the PS3 version stand out.

JaconKin,

Do you still have the image cached in your browser?

Thanks JaconKin for the images.

I recall that the R600 was the child born out of the conversations and ATI for DX10. I figured that the R600 in the 360 had some of the DX10 feature set not enabled and to keep costs down, they limited the pipelines. I guess we'll see what this upgrade/update will bring us once Vista is released.

Trying to compare framebuffers to direct feeds is retarded. You get a better picture of the two with the ones I posted:


Here's a better comparison:

PS3:

http://img297.imageshack.us/img297/5660/ps3nv2.jpg


360:

http://www.watch.impress.co.jp/game/docs/20051007/rr28.jpg

'
YOu know what you are nothing more then a PS3 fanboy. It is so obvious it isn't even funny.

qft

Trying to compare framebuffers to direct feeds is retarded. You get a better picture of the two with the ones I posted:


Here's a better comparison:

PS3:

http://img297.imageshack.us/img297/5660/ps3nv2.jpg


360:

http://www.watch.impress.co.jp/game/docs/20051007/rr28.jpg


in the above pics, it seems that the location of the road is in two different parts of a game.

the xbox is further to the side, whereas the ps3 is more towards the middle, and where the lighting textures would bring out more edge.

If the xbox was in the same part of the game, it would not be as dark, at least with the previous above posted images, the are freezing the frames on the same locations and pointing out the difference.

One of the matters I would like to bring up is the fact that it seems that the PS3 pics seems to be able to ensure possibly sharper textures in one place, the road. Yet notice the rest of the detail, the sides of the tunnel, that the PS3 must cut out of the textures.

Instead of having the black lines coming down to show the rest of the walls of the tunnel, the PS3 version has just a bright red blob. Also the PS3 version doesn't completely close the Tunnel off and has a black section going right down the middle were the roof of the tunnel would be.

So it can be deduced that in order to ensure sharper textures in one place of the game, or any game quite possibly, the RSX must give up producing textures and effects in other areas to produce sharper textures elsewhere.

Even though the 360's roads might not be as detailed as the PS3 version, at least the rest of the details are not lost.

The PS3 picture looks like it was taken after a race (tire tracks). But no racing on the 360 track.

Regardless, though, these are two different video games (6 vs 7), and this has been done to death. This thread isn't the right place for it.

I have reported all of the posts with picture comparisons as they are all off topic. It has been done to death and is not for this thread. We've seen the two systems trade off in areas of textures, resolution, and lighting, etc. depending on the title.

YOu know what you are nothing more then a PS3 fanboy. It is so obvious it isn't even funny.

Not quite rly. Who compares framebuffers to direct feeds? Tell me, and I've already stated that Xenos is powerful than RSX, while CELL is powerful than Xenon. They both their strengths and weaknesses. I'm not going to let someone post screens that doesn't tell the truth. It's unfair. Comparing framebuffers to direct feeds IS retarded, and all I've stated is that I've seen amazing post processing effects being done on CELL that I haven't seen done on 360 or even on most games for PS3.

Oh, and P.S. I've seen actual direct captures from someone playing RR6 and RR7. The same area of the game. They both look EXACTLY the same. I'd post pics, but they exceeded their bandwith and Jules already made a report to the mods.

Sorry to be a wet blanket, but there have been tons of these threads and I don't want this one to go down that road.

I have reported all of the posts with picture comparisons as they are all off topic. It has been done to death and is not for this thread. We've seen the two systems trade off in areas of textures, resolution, and lighting, etc. depending on the title.

Understood and to clean the thread up, I myself will go ahead and delete my posts that I have the pictures in. I apologize for any derailing of this thread that it might have caused. :(

In earlier posts in this thread, I stated that DMA for main memory to SPE transfers was very slow. That is a partial truth. DMA can be very fast if a large amount of data is transferred at one time but is slow if many small chunks of data is transferred. The following are the reasons.

If a large block of data is transferred, only one original large latency occurs and the remaining data can be transferred at 25.6 GB/s which is very fast.
If the size of the data to be transferred is less than one block of memory (usually 64 or 128 bytes), a long latency will occur to start the transfer and an additional latency will occur if the transfer did not start on a boundry.
If using DMA and a full block was not transferred, if the remaining part of the data is needed later, the same large latency will occur again getting that part of the data.
Normally the application must wait for the complete block to transfer prior to processing any part of the data unlike a caching system.

To illustrate these issues, an earlier post described how IBM created a realtime MPEG decoder using the concept of small block DMA transfers. Since the video buffer could not fit into the SPE memory, the SPE had to update the video buffer using DMA.

At 720p, the processor must decode a frame every 16.7 miliseconds (1/60th of a second) or else a frame may be dropped. Under normal circumstances this is not a big problem but under worst circumstances, this can be a very taxing issue.

At 720p, the video buffer is approximately 3 MB. Because of the way macroblocks relate to a raster scan video buffer, a block of data that needs to be updated in the video buffer may only be 16 bytes long. Therefore to completely write the complete video buffer, it could require approximately 175,000 DMA operations per frame. If memory has a latency of 30 nanoseconds (30 billionths of a second), the total latency time could be approximately 5.25 miliseconds or approximately 1/3 of the total of one SPE core time for processing the frame. If reads also had to be performed (occurs when motion compensation is being done), the total latency time could be double that time to approximately 2/3 of the SPE core time for the frame. If there are other conflicts with the main memory, the time can even be higher. On top of that would be the transfer time of the bytes.

That worst case may never occur and may be exagerated but it can be seen that the DMA transfers can take up a large amount of time.

Using short block transfers was not acceptable, so IBM had to redefine how the video buffer was structured so that each macroblock defined a continous area of the video buffer that was 384 bytes long. In this way, only about a maximum of 8,000 reads and/or writes would need to be performed to the video buffer creating a maximum latency of less than 1/2 milisecond in worst case conditions.

However, this creates another problem. The PPE core has to convert the non-standard video buffer to a standard raster scan video buffer and until the PPE finishes using the source video buffer, the SPE can not start processing the next frame. The PPE could convert the video buffer performing reads and writes but even with the PPEs caching system, this could take longer than desired. To reduce that time, the PPE could possibly perform a block transfer to the new video buffer (which allows the SPE to start processing the next frame) and then perform the conversion while the SPE is processing the next frame.

Why doesn't a PPE core have these latency problems? Even though the macroblocks do not define continous areas in the raster scan video buffer, the following macroblocks work on the same areas of the video bufffer. Therefore the caching system saves the old blocks for future use. Also writes are usually transparent to the processor. If continous microblocks defined areas that were totally random in the video buffer, the caching system would have problems similar to DMA (although not as bad). A caching system cannot completely eliminate the effects of latency but can greatly reduce those effects.

Using concepts such as these are not acceptable for video game programming since it wastes too much processor power both for the PPE and the SPEs. So therefore updating large objects that need to be randomly updated such as video buffers would be something that normally should not be done by an SPE in a game application.

I would say the Cell isn't all cracked out to be. A cue in logic: you can't build something inherently faster than what you are given.

Btw FNR3 looks better on 360. PS3 has the benefit that it can't do HDR+AA easily and since EA doesn't give a damn about writing PS3 suitable code, they turned off HDR and Motion Blurring trails. What you get is a 'clean' HD looking game but with even worse crowds.

RR6 and RR7 are a year apart in development in case you didn't know.

I would say the Cell isn't all cracked out to be. A cue in logic: you can't build something inherently faster than what you are given.

Btw FNR3 looks better on 360. PS3 has the benefit that it can't do HDR+AA easily and since EA doesn't give a damn about writing PS3 suitable code, they turned off HDR and Motion Blurring trails. What you get is a 'clean' HD looking game but with even worse crowds.

RR6 and RR7 are a year apart in development in case you didn't know.
wow...way to continue OT.

Get back on topic guys...

Kyser

Hey mterzich,

do you know what could be helpful in some cases is if after you post your findings and describing the technical aspects of everything, is possibly just post a simple summary in less technical talk.

I am trying to come to grips with your last post and understand it exactly what you are saying.

do you know what could be helpful in some cases is if after you post your findings and describing the technical aspects of everything, is possibly just post a simple summary in less technical talk.

I am trying to come to grips with your last post and understand it exactly what you are saying.
It is hard to summarize the post but maybe I can explain the last post a little better.

First we have to understand what latency means. Latency is the time it takes to start something to occur but after the latency (delay) occurs, everything will then occur on a regular faster basis. In the case of DRAM type memories (XDR, GDDR3, DDR2, SDRAM, etc) there is a long latency (delay) before the first byte is available but once the first byte is available, all other bytes are available very quickly. In the case of the PS3 main memory, the latency is about 100 clocks but after that, the data will stream at one byte per 1/8 of a clock.

Therefore if DMA was used to acquire 8 bytes of data it would take 100 + 8/8 = 101 clocks to get that 8 bytes of data. If the DMA were to aquire 128 bytes of data it would take 100 + 128/8 = 116 clocks. In the first case, that averages 12.625 clocks per byte and the second case averages .90625 clocks per byte.

If in the first example, the program would later aquire another 8 bytes of data that followed the original 8 bytes, it would again take 101 clocks to get the data. And if this was repeated over and over again until it finally aquired the full 128 bytes the total time required would be 101*16 = 1616 clocks. So you can see that aquiring all 128 bytes at one time is about 14x faster than aquiring the bytes 8 at a time at 16 different times. And the reason that it takes longer is that there is usually a 100 clock latency (delay) whenever memory is accessed.

The next thing that we need to understand is what is a raster scan video buffer. The easiest way to describe it is how a CRT paints the screen if it was progressive. It paints the screen starting at the upper left corner and paints the top line from left to right. When it finishes painting that line, it then paints the second line from left to right. This continues until all lines on the screen is painted. A raster scan video buffer has the information stored in the buffer sequentually in the same way it is painted on the screen. It will have the information for the first pixel of the first line at the beginning of the buffer, the information for the second pixel will follow that until the information of the 1280th pixel (at 720p) of the first line will be in the buffer. Following that will be the first pixel of the second line to be displayed. Then will be the information of the 2nd pixel of the second line and this will continue until the information for the whole 2nd line is in the buffer. The information for the 3rd line will then follow and this will continue until all the information for the 720 lines are in the buffer.

So from the above, you can see that if the programmer wants to aquire all the information of line 1, 5, 7, 100, etc., he can set the DMA to the beginning of that line and have it transfer that total line which would be fast since there is only one latency.

Now the problem is that a MPEG2 macroblock defines how a group of pixels may be represented on the screen (it could define the same color for all those pixels). Unfortunately the group that it refers to are not all in one line but instead it usually defines a square on the screen (pixels in many lines).

As an example, the macroblock may define all pixels 1 through 16 of lines 1 through 16 as having a bright red color. Since all the pixels are not continious (they are in groups of 16 on 16 different lines) in memory (only the pixels on each line is continous), 16 DMA write operations would need to be performed to change all the pixels to bright red. That causes 16 latencies (delays) giving poor performance.

So IBM said that the performance is unexceptable. How can IBM only perform one DMA write operation and still change all 16x16=256 pixels to bright red? All they did to solve the problem was make the video buffer so that pixel 1-16 of line one is followed by pixel 1-16 in line 2 and that is followed by pixel 1-16 of line 3, etc. So the whole video buffer is redifined so it is no longer a raster scan buffer but this abstract that IBM created. In this way, only only DMA operation may be required to either read or write 256 bytes that may be refered to by a macroblock greatly improving performance.

Everything is fine in the abstract, but what can they do with this mess that they created? If they sent the video buffer to the GPU and it displayed the buffer, the screen would be garbage (the GPU needs a raster scan buffer). So after a frame is generated, the PPE would need to convert that video buffer to a raster scan buffer so that the GPU could display the screen correctly.

Much of the latency problems are transparent, non-existent, or greatly reduced for the processor if a caching system is used like the PPE.

Summary

Latency or delay usually occurs everytime a new memory block of data is acquired or written.
It is much faster to read or write one large block than many small blocks of data due to latencies.
A raster scan video bufffer is a memory buffer that defines the screen the way it is painted from left to right and top to bottom.
MPEG2 macroblocks do not define a group of pixels that is continous in the raster scan video buffer so it would require many DMA reads or writes to update the video buffer with the information contained in the marcoblock.
A MPEG2 macroblock usually defines a square on the screen.
IBM redefined the video buffer so that there was a much greater chance that the marcoblock would define continous memory allowing for less DMA operations.
The redefined buffer could not be sent to the GPU since it was not a raster scan buffer. So the PPE must convert the buffer prior to sending it to the GPU.
Many of the latency problems do not exist or are greatly reduced for a PPE core due to the caching system built into the processor.

Hopefully, the above will make things a little clearer.

Note: Since I am not an expert on MPEG2 encoding, I don't know exactly how IBM redefined the video buffer but conceptually the above is correct.

i've just lost the will to live

The HD DVD group claims that HD DVD playback is the toughest application out there for the Xbox 360, and tougher than every 360 game, except for Gears of War in some parts of the game.

As I understand it, the H.264 implementation uses the programmable shaders on Xenos to assist the decompression process. I would guess that what they mean by "HD-DVD playback is completely taxing the system" is "1080p VC-1 playback is completely taxing the system".

BTW, Mterzich, you were wondering earlier about the 360s scaler, the 360 does indeed have its own scaler chip known as the ANA that is on the system board and not part of the GPU.

and the new 360 w/ HDMI will have HANA

Mterzich, although you are not a game developer, your analysis seems to be right on target.

If PS3 fan boys are still not convinced, here is an interview with John Carmack, probably the most famous game graphics developer ever that pretty much backs up what Mterzich have stated so far.

http://www.gameinformer.com/News/Story/200701/N07.0109.1737.15034.htm?Page=3

Here is what John Carmack thinks about the Cell Processor:

"I think the decision to use an asymmetric CPU by Sony was a wrong one."

the rest of that quote: "I think the decision to use an asymmetric CPU by Sony was a wrong one."
There are aspects that could make it a winning decision, but they’re not helpful to the developers. If they make the developers say that Sony is going to own the main marketplace, let’s make them develop toward this and build it this way, it would somewhat downplay the benefits of the Xbox 360 and play to the PlayStation 3’s strengths. I suspect they’re not going to overwhelmingly crush the marketplace this time, which wasn’t clear a year ago. A lot of people were thinking it’s going to be a rerun of the last generation, and it’s now looking like it might not be. I’ve been pulling for Microsoft, because I think they’ve done a better job for development support, and I think they have made somewhat smarter decisions on the platform. It’s not like the PlayStation 3 is a piece of junk or anything. I was not a fan of the PlayStation 2 and the way its architecture was set up. With the PlayStation 3, it’s not even that it’s ugly--they just took a design decision that wasn’t the best from a development standpoint.

In essence, the PS3 is a blu-ray player first.

quote from interview:

GI: Two more quick questions. Is there any game or tech that you’ve seen in the past year that’s made you say, “Wow”?

Carmack: I think Gears of War looks great. I really do. I think they did an excellent job with that. They did a lot of things really well. That’s the best-looking thing that I’ve seen in a while.

In essence, the PS3 is a blu-ray player first.
If that's what you wish to take away from it. I believe it's a powerful mulitmedia platform; music, movies, games, etc.

Now if you wanted to say "In essence the PS3 was a testbed for a new processor architecture", then I would agree. The cell has some nice features and I'm intrigued by its future iterations, another PPE, more cache per SPE, more SPEs as yeilds improve, etc. You have to lay the first first brick somewhere or you'll never build anything and Sony is looking at the cell being utilized in several devices.

I don't care which one is better then the other because I just get all the good consoles anyway. When I got my first XBOX I stopped playing my PS2. I never really take anything Kidtendo makes. This time though I may get one for gimmicks. They should try and enable 720 P for those of us with high end gaming rigs though. 480P sucks on hdtvs.

720P would be great on Wii for web browsing... But I wouldn't want to get folks used to calling the Wii 720P capable and expecting a Gears of War port! ;-)

If that's what you wish to take away from it. I believe it's a powerful mulitmedia platform; music, movies, games, etc.

A powerful multimedia platform it may be. But it certainly is not the most powerful gaming console.

And when you talk about adding pipelines and cache, you are talking about a new generation of consoles (PS4, 5, 6...), but that won't happen for at least another 5 years. You certainly can't do that to PS3 as it will make the current machines obsolete (and if you think Xbox owners are mad about future HDMI support...).

A powerful multimedia platform it may be. But it certainly is not the most powerful gaming console.

And when you talk about adding pipelines and cache, you are talking about a new generation of consoles (PS4, 5, 6...), but that won't happen for at least another 5 years. You certainly can't do that to PS3 as it will make the current machines obsolete (and if you think Xbox owners are mad about future HDMI support...).
You twisted my comment by the way. While the current Cell processor might not be the best for game developers I think it holds potential. New technology has to start some where and nobody will innately understand or fully utilize it.
Jdata said; "In essence, the PS3 is a blu-ray player first."

I said; "Now if you wanted to say "In essence the PS3 was a testbed for a new processor architecture", then I would agree. The cell has some nice features and I'm intrigued by its future iterations, another PPE, more cache per SPE, more SPEs as yeilds improve, etc. You have to lay the first first brick somewhere or you'll never build anything and Sony is looking at the cell being utilized in several devices."

I gave what I feel was a better description for the PS3 and pondered about the future of Cell processors. I made no mention of the current console or future consoles. There were several reasons why the PS3 got the cell it did and most were financial/technical. My only point was it had to start somewhere.

Not the most powerful gaming console, true but that gose to the PC of course. ;) I think it can, and will, hold its own against the 360. I don't buy into corporate hype machines, I'm sure many are sour for the pre-release statments Sony was giving and that's fine. I had realistic expectations. We can only see what the future holds. Sony was pounded for the PS2 not being developer friendly, but some great games were available for it.

Anyway, I don't want to go OT with talk of a non technical nature. Was it a mistake? Perhaps, but a seed was planted. Whether or not it bears fruit remains to be seen.

Ive done some extremely small bits of programming. After reading that interview it does make you want to give up. The people that work on these hardware platforms have their work cut out for them.

Thats about all I got to say.

It runs on magic.

Man you know all these numbers lol you guys must have graduated from MIT! Seems like he was frustrated with the new consoles when answering questions about the new consoles!

Mterzich, although you are not a game developer, your analysis seems to be right on target.

If PS3 fan boys are still not convinced, here is an interview with John Carmack, probably the most famous game graphics developer ever that pretty much backs up what Mterzich have stated so far.

http://www.gameinformer.com/News/Story/200701/N07.0109.1737.15034.htm?Page=3

Here is what John Carmack thinks about the Cell Processor:

"I think the decision to use an asymmetric CPU by Sony was a wrong one."

In the interest of completeness, although you correctly state Carmack's conclusion, you do not state his reasoning.

First, Carmack does not think that either the 360 or PS3 is ideal. He would rather have 1 fast core rather than multiple slower cores. Reality is, however, that he is faced with multiple cores because there are now limitations to making chips much faster.

Second, Carmack suggests that although he would rather have several multi-pupose cores as opposed to one multi purpose and several limited function cores, he can see how the latter could be attractive.

Third, he believes that the the PS3's approach chance of succeeding is lessened because he does not believe that the PS3 will be the dominant platform. Therefore, he does not believe that developers will develop to take advantage of the PS3's strengths. However, if a title is a PS3 exclusive, presumably, it will have been designed to take advantage of the benefits of the PS3.

If PS3 fan boys are still not convinced, here is an interview with John Carmack, probably the most famous game graphics developer ever that pretty much backs up what Mterzich have stated so far.


So in other words, and just how the entire gaming community have said, It's going to be hard to exploit CELL's potential, even thought it's powerful than Xenon, and it's going to be akin to last generation as in developers like Konami SquareEnix, Capcom, and Sony's first party/second party developers exploited Playstation 2 and made gorgeous looking games, even though it was the weakest of the three platforms.

Next time, be more logical instead of letting your fanboy instincts take over. :)

If that's what you wish to take away from it. I believe it's a powerful mulitmedia platform; music, movies, games, etc.

Now if you wanted to say "In essence the PS3 was a testbed for a new processor architecture", then I would agree. The cell has some nice features and I'm intrigued by its future iterations, another PPE, more cache per SPE, more SPEs as yeilds improve, etc. You have to lay the first first brick somewhere or you'll never build anything and Sony is looking at the cell being utilized in several devices.


Yes, that's what I have been taking it as from it from the work provided by mterzich. The Cell white paper he listed basically said it was doing do movie decoding. IMHO, you don't need much processor power to view pictures and music.

not to be an ass, but mzterich, a lot of your analysis of cell have been wrong. I've wanted to post correct info, but didnt feel like typing out a novel. A PS3 dev from the team (Ninja Theory) that is working on Heavenly Sword saw this thread and this is what he had to say:

Originally Posted by RobertR1 View Post
This is also another interesting thread on AVS with a lot of technical info which I'm sure you guys would love to argue for 10,000 posts about

http://www.avsforum.com/avs-vb/showthread.php?t=758390


Originally Posted by nAo A lot of wrong techinfo, not worth reading.That guy has no idea of what he's talking about

Read Post #18, page 1 (http://www.beyond3d.com/forum/showthread.php?t=37033)

The only reason I am doing this is because it seems that quite a few people are actually believing this incorrect information that you have been posting.

I know nAo. Very smart guy, and what they're doing with CELL and Heavenly Sword is amazing. Just one SPE is dedicated to the female's hair, lol.

The only reason I am doing this is because it seems that quite a few people are actually believing this incorrect information that you have been posting.I found it pretty interesting when I got to page 4 and read post #99, myself...

not to be an ass, but mzterich, a lot of your analysis of cell have been wrong. I've wanted to post correct info, but didnt feel like typing out a novel. A PS3 dev from the team (Ninja Theory) that is working on Heavenly Sword saw this thread and this is what he had to say:






Read Post #18, page 1 (http://www.beyond3d.com/forum/showthread.php?t=37033)

The only reason I am doing this is because it seems that quite a few people are actually believing this incorrect information that you have been posting.

The idea of a thread like this is for people to come and state why they disagree, not simply make the declaration that it is wrong and then appeal to authority. If you, or nao, disagree with what is written, please state which parts you disagree with. That would be helpful.

Thanks in advance.

People have explained already on B3D why he is wrong. Look through the pages.

People have explained already on B3D why he is wrong. Look through the pages.I already did. I saw some that agreed with nAo and plenty who disagreed. SS, DD.

The idea of a thread like this is for people to come and state why they disagree, not simply make the declaration that it is wrong and then appeal to authority.

And that is exactly what is wrong with this thread. Half-truths, assumptions, educated guesses, and opinions. I'd rather see someone defer to actual knowledge and experience rather than spew a half-baked "analysis" based on nothing but WAGs and assumptions.

Denis Dyack, says the next-generation systems are virtually identical in terms of power.

"The 360 and the PS3 are equal in power in my eyes," Dyack says. "Maybe the PS3 has more processing power. The 360 has more available memory. It's pretty much a net, net. The public perception of the PS3 was that it was much more powerful. To developers, they look even."

This is thread is turning for the worse. Never fails. Someone with no bias tries to an analyze to understand or comprehend the systems and he gets killed for it.

Others say he is wrong even though he found the information via the "net" yet hide behind other threads at other sites. The OP clearly stated his intent from the very beginning.

Either put up or shut up already.

This is thread is turning for the worse. Never fails. Someone with no bias tries to an analyze to understand or comprehend the systems and he gets killed for it.

Others say he is wrong even though he found the information via the "net" yet hide behind other threads at other sites. The OP clearly stated his intent from the very beginning.

Either put up or shut up already.
Sorry people don't agree with you assesment of the PS3
Denis Dyack, says the next-generation systems are virtually identical in terms of power.

"The 360 and the PS3 are equal in power in my eyes," Dyack says. "Maybe the PS3 has more processing power. The 360 has more available memory. It's pretty much a net, net. The public perception of the PS3 was that it was much more powerful. To developers, they look even."
I remember reading that some months ago, but I can see that as being true.

Saw this in the b3d thread:
"Conversely, the problems some may be having with Cell now will pay off with Cell2. People will have 5-6 years experience, robust librarys, and solid middleware to hit the ground running. For your parallelized code that performs nearly linear with additional SPEs you now could have an additional 20-50 SPEs to toss at the problem. Sony's strength may not be software, but they have created a bonafide platform with Cell. I expect 1st gen PS4 games to hit the road running, and running fast--as well as good BC to boot!"

Like I said things might be hard now for devs and the current Cell might not be all that great, but where it's leading is interesting.

Can everyone just agree that they both have their strengths and weaknesses as Denis Dyack mentioned?

Saw this in the b3d thread:
"Conversely, the problems some may be having with Cell now will pay off with Cell2. People will have 5-6 years experience, robust librarys, and solid middleware to hit the ground running. For your parallelized code that performs nearly linear with additional SPEs you now could have an additional 20-50 SPEs to toss at the problem. Sony's strength may not be software, but they have created a bonafide platform with Cell. I expect 1st gen PS4 games to hit the road running, and running fast--as well as good BC to boot!"

Like I said things might be hard now for devs and the current Cell might not be all that great, but where it's leading is interesting.

This is true and would be in line with sonys corporate stategy: wait for it. Wait a yr for the hardware wait a year for any games, then wait for the Cell2 when people finally learn how to use it.

From all the developer comments I've read the biggest advantage that the 360 has is that they have AWESOME support tools. They have been doing this for well over a year while sony was dinking around with the cell.

Heck go out to http://msdn.microsoft.com/directx/XNA/default.aspx and look at the kind of support Microsoft is providing to people like ME who want to just learn how to program. Its pretty robust. I can only imagine what the developers have received. This kind of stuff is afterall Microsofts biggest strength.

Sorry people don't agree with you assesment of the PS3

I remember reading that some months ago, but I can see that as being true.

Saw this in the b3d thread:
"Conversely, the problems some may be having with Cell now will pay off with Cell2. People will have 5-6 years experience, robust librarys, and solid middleware to hit the ground running. For your parallelized code that performs nearly linear with additional SPEs you now could have an additional 20-50 SPEs to toss at the problem. Sony's strength may not be software, but they have created a bonafide platform with Cell. I expect 1st gen PS4 games to hit the road running, and running fast--as well as good BC to boot!"

Like I said things might be hard now for devs and the current Cell might not be all that great, but where it's leading is interesting.
Yeah, who cares where its leading(ps4) i care about what they can do now w/ my $500 i plopped down,don't tell me about the future consoles.

This is true and would be in line with sonys corporate stategy: wait for it. Wait a yr for the hardware wait a year for any games, then wait for the Cell2 when people finally learn how to use it.

From all the developer comments I've read the biggest advantage that the 360 has is that they have AWESOME support tools. They have been doing this for well over a year while sony was dinking around with the cell.

Heck go out to http://msdn.microsoft.com/directx/XNA/default.aspx and look at the kind of support Microsoft is providing to people like ME who want to just learn how to program. Its pretty robust. I can only imagine what the developers have received.
Yes, I have read that the 360 is easy to develop for and developer support is important to have. At the same time I won't admonish Sony for bringing in a new architecture. Reading more over at b3d it looks like the cell is very capable processor. Anyway most people have their sides picked and this thread is pretty well done. Sony isn't telling anybody to wait. We can see that a new technology is just that, new.

Look at DirectX vs OpenGL. I remember when DX was a bastard child that nobody liked and now it's used for many games and OGL has fallen.
Yeah, who cares where its leading(ps4) i care about what they can do now w/ my $500 i plopped down,don't tell me about the future consoles.
I wasn't telling you anything, I don't even know who you are. I'm making a point that new technology has to grow and start out somewhere. This starting out process is usually slow esp. if there are people who don't wish to learn for it. I'm sure we'll see great improvment in 2nd gen PS3 games just like we saw for 2nd gen 360 titles, also note Gow is based on the Unreal 3 engine with 3 yrs of work put in to it.

Go read the b3d thread and it should make you feel better about the money you spent. I'm sorry you chose to purchase a console based on new architecture and now have sour grapes because things are not progressing fast enough for your taste. IIRC several titles are coming out in the next couple months, one of them being Heavenly Sword, and then you can appreciate your $500 console.

The biggest hinderence to the PS3 is the GPU, even Joker454 says this and not the new cell architecture. So I think focus is being shifted to the wrong area. It seems the cell has to be used to compensate for the lack of GPU power. I suspect Sony chose the G70 for cost and availablity reasons. IIRC the G70 was very capable during prototyping phase of the PS3 as the king of the nvidia hill were 6800s in SLi. Of course things can change pretty quickly in GPUs and thus we have a fairly unremarkable GPU by today's standards. God forbid if Sony decided to develop their own new GPU we would have even more devlopers throwing fits.

I don't have sour grapes. I am happy with my PS3 purchase, and i like resistance now better than i like gears of war. When people talk about the PS4 and how great the cell will be in the future, it sounds like they are already abandoning the PS3. I want the PS3 to be great. and hope it will. I just think Sony is making alot of mistakes.

not to be an ass, but mzterich, a lot of your analysis of cell have been wrong. I've wanted to post correct info, but didnt feel like typing out a novel. A PS3 dev from the team (Ninja Theory) that is working on Heavenly Sword saw this thread and this is what he had to say:

Read Post #18, page 1 (http://www.beyond3d.com/forum/showthread.php?t=37033)

The only reason I am doing this is because it seems that quite a few people are actually believing this incorrect information that you have been posting.

Wow... If this is what is considered as a valid argument against Mterzich, then I am sorry that I even posted here.

Nobody is arguing that PS3 is a junk. I think what Mterzich did was to peel back all of the marketing falsehoods that Sony spread that made people believe that since Cell has 8 cores instead of 3 for Xbox, that PS3 is twice the machine that Xbox 360 is.

I believe the real-life game comparison between the two consoles have demonstrated that to be false even beyond what Mterzich has stated for us.

Sony keeps saying how much better PS3 games will be in the future, but based on the architecture limitations, PS3 and Xbox 360 will always be comparable (with any difference between them being negligible to most gamers).

Sorry for jumping in. Wow, this is pretty sad.

I don't have sour grapes. I am happy with my PS3 purchase, and i like resistance now better than i like gears of war. When people talk about the PS4 and how great the cell will be in the future, it sounds like they are already abandoning the PS3. I want the PS3 to be great. and hope it will. I just think Sony is making alot of mistakes.
I'm not abandoning anything and I'm sorry if my post came across like that. People think the cell isn't all it has cracked up to be. Maybe that's true, although it's not what some are saying over at b3d, but my point was that the cell is new technology and as such it inherently takes time to bear fruit and will only be improved upon over time. The first of anything is usually a let down or not all it was hyped to be.

BTW
RFOM is a great game; I have had many 8+ hour online marathons. I look forward to many more solid titles on the PS3.

As far as mistakes from Sony. I agree they have made their fair share and the best thing to do is email, call, or post on their forums letting them know what you don't like and what you do like. It seems to 360 fans are more vocal in their demands to MS where PS3 fans seem to rabble to anybody.
Wow... If this is what is considered as a valid argument against Mterzich, then I am sorry that I even posted here.
Not to take away from Mterzich, but if you continue to read the thread you will see more specifics. I agree he has a done a good job and I thank him for his efforts.

The idea of a thread like this is for people to come and state why they disagree, not simply make the declaration that it is wrong and then appeal to authority. If you, or nao, disagree with what is written, please state which parts you disagree with. That would be helpful.

Thanks in advance.

nAo works for ninja theory which is developing the PS3 game Heavenly Sword. Dig through the MANY threads where he discusses using CELL in real world situations. Mzterich has ZERO experience with CELL and he is only speculating on many things. looking at Heavenly Sword, I'd say nAo's work speaks for itself, and if he says mzterich doesn't know what he's talking about I'm inclined to believe him.

and that thread on B3D isn't about this thread. it is actually a thread about the weaknesses of RSX. RobertR1 just pointed them over here and the posters there made a few posts commenting about this thread then went back to the main topic.

i wondered when this insightful and informative discussion would turn into a pissing contest. looks like it took about 12 pages... :(

nAo works for ninja theory which is developing the PS3 game Heavenly Sword. Dig through the MANY threads where he discusses using CELL in real world situations. Mzterich has ZERO experience with CELL and he is only speculating on many things. looking at Heavenly Sword, I'd say nAo's work speaks for itself, and if he says mzterich doesn't know what he's talking about I'm inclined to believe him.

and that thread on B3D isn't about this thread. it is actually a thread about the weaknesses of RSX. RobertR1 just pointed them over here and the posters there made a few posts commenting about this thread then went back to the main topic.

I've started reading the thread over there now, so I do appreciate you providing that link. From what I can tell so far, it's clear that nAo knows a thing or two about what he is talking about. It's also clear where his biases lay. All in all, I'm glad they aren't just deferring to his authority at that site, and are having a discussion about the issues instead.

nAo works for ninja theory which is developing the PS3 game Heavenly Sword.

Whoopee for nAo. For all you know, he could be some guy out of college who only got the job there because his uncle works there. And if you want to quote him and post some of his comments, then do so. Saying people to just go and read tens of pages of comments to get what he says is at best lazy, at worst, you don't understand what he is saying.

On the other hand, I have used direct quotes from John Carmack (where he openly compares Xbox and PS3). And if you do not know, John Carmack is probably THE BIGGEST name in game development (at least graphics wise). The guy single handly laid down the foundation for FPS genre. He has decades of experience, has worked with countless platforms, was personally responsible for many many innovations in 3D graphics. He has a friggin Emmy (how many game developers can say that?)!!!

If you don't mind, I will take John Carmack's analysis over some "dude" working on Heavenly Sword.

BTW, many of the guys at b3d only read the initial analysis done by Mzterich, which he himself corrected as he did more research on the subject, which is carefully documented in this thread.

Whoopee for nAo. For all you know, he could be some guy out of college who only got the job there because his uncle works there. And if you want to quote him and post some of his comments, then do so. Saying people to just go and read tens of pages of comments to get what he says is at best lazy, at worst, you don't understand what he is saying.
.
That only gets bits and pieces and they can be taken out of context...like the carmack quote you posted. People should do themselves a favor and read the thread rather than being lazy and asking for somebody to spoonfeed them quotes. Anyway, I'm done with this thread unless somebody would like to start posting new information of a technical nature, i.e. mterzich et al.

first of all John Carmack himself acknowledges CELL is superiour to XeCPU. In fact, it's a pretty much a fact that CELL is superiour to XeCPU. Wether or not PS3 is superiour to X360 is yet to be seen.

Second, just because he has an emmy doesnt mean crap. PS3s SIXAXIS controller won an emmy too, so does that mean PS3 is better?

third, John Carmack is very talented programmer but that's on the PC side. he's new to the console space and it's a whole different ballgame. Like I said earlier...nAo and the Ninja Theory team have heavenly sword to show what they're made of. What does Carmack have to show his worth on the console side?

Maybe you should actually read the thread if you really think he's just some nerd who's uncle got him a job there.

Second, just because he has an emmy doesnt mean crap. PS3s SIXAXIS controller won an emmy too, so does that mean PS3 is better?

.
heh, actually it didn't. Sony won for the Dualshock and Nintendo shared the award with them for the directional pad.

Damn I said I was done. Ok that's it.

first of all John Carmack himself acknowledges CELL is superiour to XeCPU. In fact, it's a pretty much a fact that CELL is superiour to XeCPU. Wether or not PS3 is superiour to X360 is yet to be seen.

There are many ways to measure performance. However, only 1 matters, and that is results. It doesn't matter what a processor is capable of doing, what matters is what it does do. It doesn't matter if one system is clocked at 10 Ghz and has 20X more cores than the other, if there isn't a noticeable difference, both systems are equal in terms of power in my eyes.

Now, I'm not saying that one system is better than the other, I am just saying that results are the only thing that matters. Give developers a few years to get the hang of both processors, until then, all arguments are pretty much pointless since we don't know what will be capable.

Most of the info about the Cell here is fanboy half-truths. There is plenty of real info on the web, IBM, Mercury and Sony are producing software, training and literature. MS owns the Xenon and it is not used outside of the 360, but the Cell has taken off to IBM's surprise outside of the PS3. While the Xenon is easier to code for the OP is misleading saying no dev will ever touch the SPUS, Insomniac has already shown us how they have used them for Resistance. Of course first gen games probably won't use the SPUs, just like the first gen 360 games didn't use more than one core. We got games like God of War from the PS2 and it's insane architecture, we will see better results from the Cell+RSX.

Sorry to break up the Xbox fanboy circle jerk, if you guy really are interested in the cell and not just bashing the PS3 please check the web for tons of info.

http://www.research.ibm.com/cell/
http://arstechnica.com/articles/paedia/cpu/cell-1.ars
http://arstechnica.com/articles/paedia/cpu/cell-2.ars
http://www.blachford.info/computer/Cell/Cell0_v2.html
http://www.mc.com/cell/

Most of the info about the Cell here is fanboy half-truths. There is plenty of real info on the web, IBM, Mercury and Sony are producing software, training and literature. MS owns the Xenon and it is not used outside of the 360, but the Cell has taken off to IBM's surprise outside of the PS3.

To IBM's surprise? The Cell Broadband Engine was not created with the PS3 in mind. Sony wanted to use Cell in the PS3, but IBM mainly wanted it for servers. Heck, I bet Sony isn't very happy with how it came out. They originally wanted the Cell to do ALL the processing in the PS3 as a CPU/GPU combo chip. It is still a good processor, just not nearly as good as Sony says/expected.

Developers are going to use the SPEs, there really isn't any denying it. If they don't, the PS3 will never catch up to the Xbox 360 in terms of graphics. The argument is whether or not they will use it effectively.

Frankly, the Cell is just potential at this point. Like I said earlier, It doesn't matter if the Cell has 20 SPEs, if they are not effectively used, all you're paying for is potential.

Now, don't get me wrong, I don't think one system is better than the other in terms of graphics. They both have their advantages and disadvantages, just don't except one of them to blow the other away. This is been said many times, if you're shown a picture of a random game, you won't be able to tell which system it is for. Just be happy with the games.

Not to bash on Sony, but I think they are responsible for this...They tend to mislead their fans. RSX was touted as a monstrous card that can reach performances of 1.8 Teraflops...The X1950XTX and 7950 GX2, cards which completely overpower the 7800GTX can barely achieve 600 Gigaflops.

I thought it was well established that GPU wise, 360 wins, and CPU wise, PS3 wins. There you go. Both have their advantage and disadvantage.


Sorry for going OT, but Jetrii. I remember you and I having an argument a month ago with you saying that 360 would reach 10 million? :D

Most of the info about the Cell here is fanboy half-truths. There is plenty of real info on the web, IBM, Mercury and Sony are producing software, training and literature. MS owns the Xenon and it is not used outside of the 360, but the Cell has taken off to IBM's surprise outside of the PS3. While the Xenon is easier to code for the OP is misleading saying no dev will ever touch the SPUS, Insomniac has already shown us how they have used them for Resistance. Of course first gen games probably won't use the SPUs, just like the first gen 360 games didn't use more than one core. We got games like God of War from the PS2 and it's insane architecture, we will see better results from the Cell+RSX.

Sorry to break up the Xbox fanboy circle jerk, if you guy really are interested in the cell and not just bashing the PS3 please check the web for tons of info.

http://www.research.ibm.com/cell/
http://arstechnica.com/articles/paedia/cpu/cell-1.ars
http://arstechnica.com/articles/paedia/cpu/cell-2.ars
http://www.blachford.info/computer/Cell/Cell0_v2.html
http://www.mc.com/cell/

Give it up, they don't understand the technical facts. They will believe it once they see on their screens. It will come...starting very soon. Microsoft is nervous, it's why they started calling Sony with all sorts of names.

Sorry for going OT, but Jetrii. I remember you and I having an argument a month ago with you saying that 360 would reach 10 million? :D

Yeap. From the looks of it, they passed the 10 million mark :)



Give it up, they don't understand the technical facts. They will believe it once they see on their screens. It will come...starting very soon. Microsoft is nervous, it's why they started calling Sony with all sorts of names.

No offense but if you really understood the facts, you would know why people are questioning the power of the Cell. No one (aside from blind fanboys) is saying that the Cell processor is not powerful. It is very powerful at certain things. That is the key, certain things. Game code is very chaotic, it will be a real pain to effectively use all the SPEs. I can't say I have as much experience as developers working on console games but I have done my share of C++/ASM programming. I can't begin to imagine the pain I would go through if I had to program for the cell. I would eventually get it, but it would be a pain. I don't think any of the people who actually understand the technical info is saying that the xbox 360 is "better" than the PS3, I just don't think the PS3 is capable of anything that the Xbox 360 can't closely match...But like you said, lets wait.

Dralt, I have a question for you. What do you expect out of the PS3 graphic wise? Could you please give me a comparison of what you think the PS3 and Xbox 360 are capable of?

Yeap. From the looks of it, they passed the 10 million mark


Actually, they didn't. :)


Anyway, I think that 360's CPU is actually very powerful. If the rumor about Ubi Montreal is doing with SC5, and I do believe it's true, since the info got leaked and was quickly deleted by a mod on Gamespot's forum, then boy oh boy. The animation system is going to be SICK.

Actually, they didn't. :)

Do you have a link to support this? I recall Microsoft saying they sold over 10 million units in several articles.

To quote Bill Gates:
"Games for Windows is a growing market. But of course, while leading platform, our 2nd is Xbox 360. We've become the leader in this next gen. We've sold 10.4 million consoles in 37 countries."

They sold 10.4 million units to RETAILERS. They didn't actually sell it to customers. With the 360 at 4.5 million units in the U.S., they'd be lucky to have 7.5 million sold worldwide.


Here's what I think will definately prove 360's CPU is also a beast:

http://forums.ubi.com/eve/forums/a/tpc/f/2771048463/m/7041028584/p/21

They sold 10.4 million units to RETAILERS. They didn't actually sell it to customers. With the 360 at 4.5 million units in the U.S., they'd be lucky to have 7.5 million sold worldwide.

Microsoft has actually been very forward about shipping vs selling. If you search back, they announced when they shipped 6 million units and later when they sold 6 million units.


Here's what I think will definately prove 360's CPU is also a beast:

http://forums.ubi.com/eve/forums/a/tpc/f/2771048463/m/7041028584/p/21


Yeap, there is no arguing that both processors are pretty powerful.

That's nice, but too bad. They missed their mark. Still did good, though.

That's nice, but too bad. They missed their mark. Still did good, though.

Well, my good friend Gates is a bit more credible than you in this area. You can speculate that they didn't sell 10 million units, but until proven otherwise in official reports, I think I'll go with him. Speaking of which, I am very close to disproving the aged theory that the earth is round.

http://www.gamesindustry.biz/content_page.php?aid=22123

"I'll defer my comments. "Shipped," for the purpose of what they're talking about, is irrelevant. "Shipped" means it's left the factory. What we're talking about is "sold." "Sold" means that we're a wholesaler of hardware and we sell it to a retailer, and that's the important criteria."

Ahem.

http://www.gamesindustry.biz/content_page.php?aid=22123

"I'll defer my comments. "Shipped," for the purpose of what they're talking about, is irrelevant. "Shipped" means it's left the factory. What we're talking about is "sold." "Sold" means that we're a wholesaler of hardware and we sell it to a retailer, and that's the important criteria."

Ahem.

I stand corrected about the 10.4 million. Thanks for the article. However, If I recall correctly, Walmart does not provide NPD with data about it's sales. Even if you add all the available data together, it still passes 8/9 million units.

Did you read the article from the Ubi board? If the game testers couldn't find Sam performing the same canned animation twice when jumping, then yeah. I would really like someone here who knows a lot about the PS3 and 360's CPU on animation techniques. I think that should be the most important aspect when multi threading games.

No offense but if you really understood the facts, you would know why people are questioning the power of the Cell. No one (aside from blind fanboys) is saying that the Cell processor is not powerful. It is very powerful at certain things. That is the key, certain things. Game code is very chaotic, it will be a real pain to effectively use all the SPEs. I can't say I have as much experience as developers working on console games but I have done my share of C++/ASM programming. I can't begin to imagine the pain I would go through if I had to program for the cell. I would eventually get it, but it would be a pain. I don't think any of the people who actually understand the technical info is saying that the xbox 360 is "better" than the PS3, I just don't think the PS3 is capable of anything that the Xbox 360 can't closely match...But like you said, lets wait.

Dralt, I have a question for you. What do you expect out of the PS3 graphic wise? Could you please give me a comparison of what you think the PS3 and Xbox 360 are capable of?

No offense, but I know at least much as you do on this topic.

The deal is this one:

A general purpose processing architecture (Xbox 360) vs. a processing architecture targeting multimedia applications specifically (PS3)

False sharing vs. no sharing

So, I expect some developers, who like to go down to the metal, will show the PS3 is superior to the 360 by the same margin the Xbox was superior to the PS2.

Now, I also expect a large number of developers to never "find" the power of the PS3, because they will use middleware solutions in an attempt to find it.

No offense, but I know at least much as you do on this topic.

The deal is this one:

A general purpose processing architecture (Xbox 360) vs. a processing architecture targeting multimedia applications specifically (PS3)

False sharing vs. no sharing

So, I expect some developers, who like to go down to the metal, will show the PS3 is superior to the 360 by the same margin the Xbox was superior to the PS2.

Now, I also expect a large number of developers to never "find" the power of the PS3, because they will use middleware solutions in an attempt to find it.

Well, looks like we have to wait until the games reflect it :)

http://www.gamesindustry.biz/content_page.php?aid=22123

"I'll defer my comments. "Shipped," for the purpose of what they're talking about, is irrelevant. "Shipped" means it's left the factory. What we're talking about is "sold." "Sold" means that we're a wholesaler of hardware and we sell it to a retailer, and that's the important criteria."

Ahem.

Peter Moore:

They should also keep a focus on their friends to the north in Kyoto

Someone should tell Pete Kyoto is actually to the South of Tokyo. Hopefully, his management skills are superior to his knowledge of geography.

Why are there so many opinions and so few facts here?

This is getting way off topic here guys and is turning into a console wars issue.

Keep this on topic, it was a great discussion.

If it keeps going downhill I'll be forced to close it.

Kyser

I think that the games that come out for each system will ultimately tell the tale of real world power for both proccessors. In 3 years we will know. The 360's proccessor is not maxed out yet, and th ePS3's is obviously not either.

The consensus seems to be that the Xenos is more straightforward and easy to program for, and the cell is largely unknown but holds great potential.

Porting 360 Game Applications to a PC

Since the 360 has a general purpose 3-core processor, it also has one more advantage that has not yet been mentioned. Any game application that has been developed for the xenon processor can be easily be ported to run on a 1, 2, or 4 core processor. From a game developer that also wants to support the game application on the PC, this is very important since most currently available PC processors are 2 and 4 core processors and it should be fairly easy to port to those processors.

Therefore any game application that was developed for the 360 could be easily ported to the PC by just changing any needed library calls and then assigning the different threads to different cores (or software threads for single core processors) during runtime. Since any race condition should have already been solved previously with the 3-core 360 implementation, there shouldn't be any major problems during the port. Obviously if a single or dual core processor does not have enough processor power, performance would be hurt. Also it may be risky to use all 4 cores of a quad processor (when porting from the 3-core xenon processor) due to possible race conditions so possibly only 3 of the 4 cores may be used.

Cell Processor in Massively Parallel Computers

IBM has developed a blade with dual cell processors (2 PPE cores and 16 SPE cores) for use in their massively parallel computers. The advantage of doing this is to greatly reduce the cost but these blades are primarily for specialized use where there would be a need for massive amount of parallel vector operations but at greatly reduced price. How well these blades will sell is anyones guess since the applications would need to be re-coded to use the cell processors and generally those users are generally well funded (wheather, nuclear research, etc.). Normally massively parallel computers contain general purpose processors with common main memory and can be used in many aspects of business, government, graphics development, research, etc. The following is one example of a massively parallel computer from Silicon Graphics.

http://www.sgi.com/company_info/newsroom/virt_presskit/nab05/downloads/prism_for_media.pdf

Some to the things that you will notice about this computer system.

Maximum of 128 processors (256 cores) using Intel's Itanium 2 Processor
Maximum of 4,000 GB main memory
Maximum main memory bandwidth of 1,280 GB/s
Maximum of 64 memory buses (20 GB/s each)

The following is the latest Intel Itanium 2 Processor.

http://en.wikipedia.org/wiki/Montecito_%28processor%29

This processor primarily relies on L3 cache instead of L2 cache to give it the performance that is needed. I suspect that it was done this way to reduce the power requirement and to reduce the heat produced. The following are some of the interesting features of that processor.

1.72 billion transistors on 24 MB L3 cache version
Maximum 24 MB L3 cache using 1.550 billion transistors for L3 cache
Cost per processor (2 cores) in quanities for a 24 MB L3 cache version is $3,692

As you can see, the emphasis for the computer system and processor is placed on memory speed with large L3 cache and multiple memory buses.

first of all John Carmack himself acknowledges CELL is superiour to XeCPU. In fact, it's a pretty much a fact that CELL is superiour to XeCPU. Wether or not PS3 is superiour to X360 is yet to be seen.

Second, just because he has an emmy doesnt mean crap. PS3s SIXAXIS controller won an emmy too, so does that mean PS3 is better?

third, John Carmack is very talented programmer but that's on the PC side. he's new to the console space and it's a whole different ballgame. Like I said earlier...nAo and the Ninja Theory team have heavenly sword to show what they're made of. What does Carmack have to show his worth on the console side?

Maybe you should actually read the thread if you really think he's just some nerd who's uncle got him a job there.


I hesitate to post this, because I don't want it to be taken as a general slam on the PS3. Instead, I'm only mentioning this because it shows how some developers feel about the Cell.

Beyond the above mentioned John Carmack, we now also have Valve Software's Gabe Newell chimming in on the PS3, saying the following:

“The PS3 is a total disaster on so many levels, I think It’s really clear that Sony lost track of what customers and what developers wanted. I’d say, even at this late date, they should just cancel it and do a ‘do over’. Just say, ‘This was a horrible disaster and we’re sorry and we’re going to stop selling this and stop trying to convince people to develop for it."
http://www.cinemablend.com/games/Gabe-Newell-Rips-PS3-A-New-One-2362.html

Which sounds to me like he also doesn't particularly care to program for the Cell. I suppose there could be other interpretations, although that sounds like the most likely.

And that's why I'm hesitatant to just trust nAo, and why I would rather see the pro's and con's worked over here. It looks like there is a split of opinion in the dev's world about the Cell.

Keep this on topic, it was a great discussion.

If it keeps going downhill I'll be forced to close it.

Kyser
I think you should it's gotten a bit long in the tooth.

For people that do not have a technical background, the following is an analogy.

Assume that someone has all the parts to build a car in his garage (main memory). That person could build the car from an instruction manual but it would take a long time (single core PPE application).

If he gets two of his friends to help him and they were all working in the garage with him, they may be able to get the car built in 1/3 the time. (3-core 360 application). The main issue with having three people working on the car at the same time is that everything has to be better organized so that everyone is working in an orderly fashion and they are not falling over each other.

Now if that person got 6 of his friends and he put each of them in a different 4'x2' closet (256 KB SPE memory) and he is the only person allowed to be in a garage. He would then have to bring parts to each of his friends with a page out of the instruction booklet and tell them to assemble the parts. Only parts that could fit in the closet could be sent to the friends working in closets. When the assembled parts are ready, he would go and get the assembled parts and bring new parts and instructions for assembly to the closet. Also parts that do not take very long to assemble should be assembled in the garage since it may take longer to organize and bring and retrieve those parts to and from the closet than assembling them in the garage. Anything that could not be assembled in the closet, he would have to assemble it in the garage. He would also have to assemble all the assembled parts in the garage to produce the car (a cell application using 1 PPE core and 6 SPEs). If a another car (next video frame) was to be built, the completed last car could would have to be taken from the garage (sent to the gpu) and new parts aquired (user controls) and the person in the garage would need to organize the parts so that tasks could be performed.

Even though the final example has 7 people working on the car at the same time, it will probably not be finished 7x as fast since people in the closets are limited to what they can work on due to space restrictions (256 KB memory) and transportation time for small assemblies whereas the person working in the garage has to do so much work (managing everything in an orderly fashion, bring parts and pickup parts to/from people in the closets, assemble all large parts, assemble all low manpower parts, as well as performing final assembly for the car). In this case, the people in the closets may be idle since the main person is too busy performing too many tasks. The main person is the limiting factor.

Great thread mterzich. Are you as familiar with GPUs as you are with CPUs?

Please just stop mterzich, it's pathetic. You try to come off as technical and fair but it's plain to everyone not already drinking the same koolaide that you have made a thread to cater to fanboys. You have done nothing but to pump up the 360 CPU and denegrade the Cell. You would get raped at a technical forum like Beyond3D with real developers. Why this wasn't locked already is beyond me, it's mostly BS.

Please just stop mterzich, it's pathetic. You try to come off as technical and fair but it's plain to everyone not already drinking the same koolaide that you have made a thread to cater to fanboys. You have done nothing but to pump up the 360 CPU and denegrade the Cell. You would get raped at a technical forum like Beyond3D with real developers. Why this wasn't locked already is beyond me, it's mostly BS.

Is this your way of saying, "I don't know why you're wrong, but I don't like what you're saying, so please stop saying it"? :)

Is this your way of saying, "I don't know why you're wrong, but I don't like what you're saying, so please stop saying it"? :)

Don't bother replying, just ignore him. It is his way of trying to start a flame war to close this great thread.

Great thread mterzich. Are you as familiar with GPUs as you are with CPUs?
I know very little about GPUs other than they are very specialized processors. In the case of a GPU, the more specialized the processor is, the better. This is because it performs repetive specialized tasks and the better the hardware is designed for those type of tasks, the beter it will perform.

Don't bother replying, just ignore him. It is his way of trying to start a flame war to close this great thread.


Exactly right Jetrii. BTW - I've been meaning to say "great sig" for quite some time.

Exactly right Jetrii. BTW - I've been meaning to say "great sig" for quite some time.

Thanks :) Follow it well.

For people that do not have a technical background, the following is an analogy.

Assume that someone has all the parts to build a car in his garage (main memory). That person could build the car from an instruction manual but it would take a long time (single core PPE application).

If he gets two of his friends to help him and they were all working in the garage with him, they may be able to get the car built in 1/3 the time. (3-core 360 application). The main issue with having three people working on the car at the same time is that everything has to be better organized so that everyone is working in an orderly fashion and they are not falling over each other.

Now if that person got 6 of his friends and he put each of them in a different 4'x2' closet (256 KB SPE memory) and he is the only person allowed to be in a garage. He would then have to bring parts to each of his friends with a page out of the instruction booklet and tell them to assemble the parts. Only parts that could fit in the closet could be sent to the friends working in closets. When the assembled parts are ready, he would go and get the assembled parts and bring new parts and instructions for assembly to the closet. Also parts that do not take very long to assemble should be assembled in the garage since it may take longer to organize and bring and retrieve those parts to and from the closet than assembling them in the garage. Anything that could not be assembled in the closet, he would have to assemble it in the garage. He would also have to assemble all the assembled parts in the garage to produce the car (a cell application using 1 PPE core and 6 SPEs). If a another car (next video frame) was to be built, the completed last car could would have to be taken from the garage (sent to the gpu) and new parts aquired (user controls) and the person in the garage would need to organize the parts so that tasks could be performed.

Even though the final example has 7 people working on the car at the same time, it will probably not be finished 7x as fast since people in the closets are limited to what they can work on due to space restrictions (256 KB memory) and transportation time for small assemblies whereas the person working in the garage has to do so much work (managing everything in an orderly fashion, bring parts and pickup parts to/from people in the closets, assemble all large parts, assemble all low manpower parts, as well as performing final assembly for the car). In this case, the people in the closets may be idle since the main person is too busy performing too many tasks. The main person is the limiting factor.

What happens when 3 main persons need to use a single wrench at the same time?

What happens when 3 main persons need to use a single wrench at the same time?

Why would they only have 1 wrench? They each have 2 arms(hardware threads), they could use 6 wrenches if they really need to.

What happens when 3 main persons need to use a single wrench at the same time?
The analogy is not perfect. A more important issue is that tasks must be organized and scheduled in orderly fashion (ex. someone can not assemble an engine if the assembly of pistons, carbarator, etc. have not yet been performed).

This is why scheduling of tasks by the main core is so important and one of the reasons why a 3-core application may not perform at 3x the speed of a single core application. All this scheduling and organization increases complexity and development cost. A program must define exact tasks in advance and if the wrong order is scheduled (unlike a human who can change his mind when he realizes that he should be doing something else), a core may remain idle for an extended period of time or even crash if the dependencies are not available. Dependencies are one of the most difficult problems to solve when developing a multi-core application.

mterzich, I found some information about the GPUs on beyond3D.

GPU Transistor Count
PS3 - RSX transistor count: 300.2 million transistors
Xbox 360 - Xenos transistor count: 337 million (232 million parent die+105 million EDRAM daughter die)

GPU clock
Xbox 360 - Xenos clocked at 500 Mhz
PS3 - RSX clocked at 500 MHz

GPU video memory
Xbox 360 - Xenos: 512 MB of 700 Mhz GDDR3 VRAM on a 128-bit bus
Xbox 360 - Xenos: 10 MB daughter Embedded DRAM as framebuffer (32GB/s bus, multiplied by 8 thanks to multisampling unpacking for an effective bandwidth of 256 MB/s, the internal eDRAM bandwidth)
PS3 - RSX: 256 MB GDDR3 VRAM clocked at 700 Mhz on a 128-bit bus
PS3 - RSX: 256 MB of Rambus XDR DRAM via Cell (with latency penalty)

Triangle Setup
Xbox 360 - 500 Million Triangles/sec
PS3 - 250 Million Triangles/sec

Vertex Shader Processing
Xbox 360 - 6.0 Billion Vertices/sec (using all 48 Unified Pipelines)
Xbox 360 - 2.0 Billion Vertices/sec (using only 16 of the 48 Unified Pipelines)
Xbox 360 - 1.5 Billion Vertices/sec (using only 12 of the 48 Unified Pipelines)
Xbox 360 - 1.0 Billion Vertices/sec (using only 8 of the 48 Unified Pipelines)
PS3 - 1.1 Billion Vertices/sec (if all 8 Vertex Pipelines remain)
PS3 - 0.825 Billion Vertices/sec (if downgraded to 6 Vertex Pipelines)

Filtered Texture Fetch
Xbox 360 - 8.0 Billion Texels/sec
PS3 - 13.2 Billion Texels/sec (if all 24 Pixel Pipelines remain)
PS3 - 11.0 Billion Texels/sec (if downgraded to 20 Pixel Pipelines)

Vertex Texture Fetch
Xbox 360 - 8.0 Billion Texels/sec
PS3 - 4.4 Billion Texels/sec (if all 8 Vertex Pipelines remain)
PS3 - 3.3 Billion Texels/sec (if downgraded to 6 Vertex Pipelines)

Pixel Shader Processing with 16 Filtered Texels Per Cycle (Pixel ALU x Clock)
Xbox 360 - 24.0 Billion Pixels/sec (using all 48 Unified Pipelines)
Xbox 360 - 20.0 Billion Pixels/sec (using 40 of the 48 Unified Pipelines)
Xbox 360 - 18.0 Billion Pixels/sec (using 36 of the 48 Unified Pipelines)
Xbox 360 - 16.0 Billion Pixels/sec (using 32 of the 48 Unified Pipelines)
PS3 - 17.6 Billion Pixels/sec (if all 24 Pixel Pipelines remain)
PS3 - 13.2 Billion Pixels/sec (if downgraded to 20 Pixel Pipelines)

Pixel Shader Processing without Textures (Pixel ALU x Clock)
Xbox 360 - 24.0 Billion Pixels/sec (using all 48 Unified Pipelines)
Xbox 360 - 20.0 Billion Pixels/sec (using 40 of the 48 Unified Pipelines)
Xbox 360 - 18.0 Billion Pixels/sec (using 36 of the 48 Unified Pipelines)
Xbox 360 - 16.0 Billion Pixels/sec (using 32 of the 48 Unified Pipelines)
PS3 - 26.4 Billion Pixels/sec (if all 24 Pixel Pipelines remain)
PS3 - 22.0 Billion Pixels/sec (if downgraded to 20 Pixel Pipelines)

Multisampled Fill Rate
Xbox 360 - 16.0 Billion Samples/sec (8 ROPS x 4 Samples x 500MHz)
PS3 - 8.0 Billion Samples/sec (8 ROPS x 2 Samples x 500MHz)

Pixel Fill Rate with 4x Multisampled Anti-Aliasing
Xbox 360 - 4.0 Billion Pixels/sec (8 ROPS x 4 Samples x 500MHz / 4)
PS3 - 2.0 Billion Pixels/sec (8 ROPS x 2 Samples x 500MHz / 4)

Pixel Fill Rate without Anti-Aliasing
Xbox 360 - 4.0 Billion Pixels/sec (8 ROPS x 500MHz)
PS3 - 4.0 Billion Pixels/sec (8 ROPS x 500MHz)

Frame Buffer Bandwidth
Xbox 360 - 256.0 GB/sec (dedicated for frame buffer rendering)
PS3 - 22.4 GB/sec (shared with other graphics data: textures and vertices)
PS3 - 12.4 GB/sec (with 10.0 GB/sec subtracted for textures and vertices)
PS3 - 10.0 GB/sec (with 12.4 GB/sec subtracted for textures and vertices)

Texture/Vertex Memory Bandwidth
Xbox 360 - 22.4 GB/sec (shared with CPU)
Xbox 360 - 14.4 GB/sec (with 8.0 GB/sec subtracted for CPU)
Xbox 360 - 12.4 GB/sec (with 10.0 GB/sec subtracted for CPU)
PS3 - 22.4 GB/sec (shared with frame buffer)
PS3 - 12.4 GB/sec (with 10.0 GB/sec subtracted for frame buffer)
PS3 - 10.0 GB/sec (with 12.4 GB/sec subtracted for frame buffer)
PS3 - additional 20.0 GB/sec when reading from XDR memory (with latency penalty)

Shader Model
Xbox 360 - Shader Model 3.0+ / Unified Shader Architecture
PS3 - Shader Model 3.0 / Discrete Shader Architecture

The analogy is not perfect. A more important issue is that tasks must be organized and scheduled in orderly fashion (ex. someone can not assemble an engine if the assembly of pistons, carbarator, etc. have not yet been performed).

This is why scheduling of tasks by the main core is so important and one of the reasons why a 3-core application may not perform at 3x the speed of a single core application. All this scheduling and organization increases complexity and development cost. A program must define exact tasks in advance and if the wrong order is scheduled (unlike a human who can change his mind when he realizes that he should be doing something else), a core may remain idle for an extended period of time or even crash if the dependencies are not available. Dependencies are one of the most difficult problems to solve when developing a multi-core application.

I'm not sure you answered my question... :D

I'll tell you my answer...2 cores remain idle while 1 core uses the wrench.

Is that what you meant?

I'm not sure you answered my question... :D

I'll tell you my answer...2 cores remain idle while 1 core uses the wrench.

Is that what you meant?
No that is not quite correct. In the case of processor cores, each of them always has the tools (execution units) available at all times. With good programming, the only thing that causes a processor to become idle is waiting on dependencies to be completed so that task for a core can be started.

Sometimes, the developer will have one core control itself as well as the other cores by telling a core to start executing certain routines (instructions) when all input data (parts) are ready to be used by the specific core. Another concept uses what is called semaphores for waking up and halting processor cores depending on whether the input data is available.

Jetrii,

The information that you found on the GPUs are interesting. From a paper specification it appears that the 360 GPU is more powerful but without fully understanding how the harware is put together and how a gpu is programmed, it would be pure unsubstantiated speculation to indicate whether one is better than the other.

Since my knowledge of gpus are very limited, I would not be able to draw any conclusions from those specifications. Speculating on that without the proper knowledge would be no different than Sony claiming that the cell processor is so much more powerful than the xenon processor since it has 3 times as many cores and over 2 times the GFLOPS.

I just posted those numbers in case you were interested, I know that they are not an accurate measurement of performance.

I know very little about GPUs other than they are very specialized processors. In the case of a GPU, the more specialized the processor is, the better. This is because it performs repetive specialized tasks and the better the hardware is designed for those type of tasks, the beter it will perform.

You don't know much about CELL either.

You don't know much about CELL either.


Boy, that was real constructive! How about informing him, me and others instead of giving out quick and short comments that mean nothing.

We're waiting.

No that is not quite correct. In the case of processor cores, each of them always has the tools (execution units) available at all times. With good programming, the only thing that causes a processor to become idle is waiting on dependencies to be completed so that task for a core can be started.

I quote:

"Xenon is equipped with a 1 MiB Level 2 cache on-die running at half CPU clock speed. This cache is shared amongst the three CPU cores."

So, what you say is not quite correct. :)

I also suggest you look up cache line invalidation and false sharing.

Sometimes, the developer will have one core control itself as well as the other cores by telling a core to start executing certain routines (instructions) when all input data (parts) are ready to be used by the specific core. Another concept uses what is called semaphores for waking up and halting processor cores depending on whether the input data is available.

You keep mixing concepts from very different abstraction levels. From processor core to semaphore, there is quite a gap, for instance.
Cache line invalidation and false sharing do happen and it's not a matter of threading model or thread implementation.
The only way you can avoid them is by wasting cache space.

I quote:

"Xenon is equipped with a 1 MiB Level 2 cache on-die running at half CPU clock speed. This cache is shared amongst the three CPU cores."

So, what you say is not quite correct. :)

I also suggest you look up cache line invalidation and false sharing.



You keep mixing concepts from very different abstraction levels. From processor core to semaphore, there is quite a gap, for instance.
Cache line invalidation and false sharing do happen and it's not a matter of threading model or thread implementation.
The only way you can avoid them is by wasting cache space.
I'm not sure what the point is that you are trying to make. I tried to make an analogy that could be understood by a non-technical person and therefore not be too complex. This analogy or any anology is not exact.

I could have made the analogy more complicated by having 3 robots in the garage working from prewritten instructions to represent the xenon processor. It would also stated that the robots could only work on specific tasks and could not directly assist another robot (such as lifting and installing the engine in the car). May other completities would also have to be stated to get the analogy closer.

As far a the cell processor anology, again robots would be put in the closet and may have to be replaced with a different robot if that robot did not have the skills to perform a certain task (code in the SPE cannot process that specific data). Also a robot (or SPE) could possibly initiate the transportion of parts (data) to and from the closet but this again would create more complexity in the analogy.

I did not mention that tempory stalls can be caused by cache misses (due to needed memory not being in the cache or cache invalidation due to DMA transfers) since it is relatively unimportant in the overall analogy. I agree that I also mix up software and hardware when describing the analogy since the fit is not perfect and that was the best that I could come up with.

If you don't like my analogy, please feel free to create one that is more exact.

I think a better analogy for the Cell precessor is a racing pit crew. Each SPE is assigned an individual task very quickly.



For people that do not have a technical background, the following is an analogy.

Assume that someone has all the parts to build a car in his garage (main memory). That person could build the car from an instruction manual but it would take a long time (single core PPE application).

If he gets two of his friends to help him and they were all working in the garage with him, they may be able to get the car built in 1/3 the time. (3-core 360 application). The main issue with having three people working on the car at the same time is that everything has to be better organized so that everyone is working in an orderly fashion and they are not falling over each other.

Now if that person got 6 of his friends and he put each of them in a different 4'x2' closet (256 KB SPE memory) and he is the only person allowed to be in a garage. He would then have to bring parts to each of his friends with a page out of the instruction booklet and tell them to assemble the parts. Only parts that could fit in the closet could be sent to the friends working in closets. When the assembled parts are ready, he would go and get the assembled parts and bring new parts and instructions for assembly to the closet. Also parts that do not take very long to assemble should be assembled in the garage since it may take longer to organize and bring and retrieve those parts to and from the closet than assembling them in the garage. Anything that could not be assembled in the closet, he would have to assemble it in the garage. He would also have to assemble all the assembled parts in the garage to produce the car (a cell application using 1 PPE core and 6 SPEs). If a another car (next video frame) was to be built, the completed last car could would have to be taken from the garage (sent to the gpu) and new parts aquired (user controls) and the person in the garage would need to organize the parts so that tasks could be performed.

Even though the final example has 7 people working on the car at the same time, it will probably not be finished 7x as fast since people in the closets are limited to what they can work on due to space restrictions (256 KB memory) and transportation time for small assemblies whereas the person working in the garage has to do so much work (managing everything in an orderly fashion, bring parts and pickup parts to/from people in the closets, assemble all large parts, assemble all low manpower parts, as well as performing final assembly for the car). In this case, the people in the closets may be idle since the main person is too busy performing too many tasks. The main person is the limiting factor.

It seems like the 360 is a winner in every aspect, Is there a situation where the PS3 hardware has an advantage? BTW Im no techie or programmer just curious so I can wow my friends during gatherings!


Yes, the PS3 has a hardware advantage when it comes to raw CPU power (more than twice as much). However it will be awhile until we see games created that fully utilize that extra CPU power it has. Ports or multiplatform games that appear on the PC, 360, Wii, and PS3/Ps2 will look best on the 360 since its hardware is most similar to a PC's and it is easier to program so Ps3 optimizations will be limited (Also the Wii versions of those games might be the most fun to play as well without graphics being an issue). My prediction is that the 360 has almost peaked when it comes to graphics with the game "Gears of War" and I see Halo 3 trumping it once that is released. However the Ps3's future game exclusives that come out in a few years like God of War 3 for example may look more advanced due to being able to use more CPU power more efficiently.All that is speculation and in the meantime both consoles are producing excellent games with interesting titles yet to be released. I am not surprised this discussion has turned out the way it has since it was posted in the X-Box section of this forum..

This thread is truly amazing. I think it could be featured prominently in a study of various mental illnesses.

Nothing personal to anyone. ;)

This thread is truly amazing. I think it could be featured prominently in a study of various mental illnesses.

Nothing personal to anyone. ;)

:D Agreed!

Yes, the PS3 has a hardware advantage when it comes to raw CPU power (more than twice as much). However it will be awhile until we see games created that fully utilize that extra CPU power it has. Ports or multiplatform games that appear on the PC, 360, Wii, and PS3/Ps2 will look best on the 360 since its hardware is most similar to a PC's and it is easier to program so Ps3 optimizations will be limited (Also the Wii versions of those games might be the most fun to play as well without graphics being an issue). My prediction is that the 360 has almost peaked when it comes to graphics with the game "Gears of War" and I see Halo 3 trumping it once that is released. However the Ps3's future game exclusives that come out in a few years like God of War 3 for example may look more advanced due to being able to use more CPU power more efficiently.All that is speculation and in the meantime both consoles are producing excellent games with interesting titles yet to be released. I am not surprised this discussion has turned out the way it has since it was posted in the X-Box section of this forum..


your just using a theory that you thought up. the ps3 does not necassarly have more RAW power it only does that in a given situation. Also the 360 has a much more effective and powerful GPU and that is where a HUGE amount of what a game looks like . Epic openly admits that gears doesnt touch what the xbox can do and knowing what they know now they could do better.

This thread is truly amazing. I think it could be featured prominently in a study of various mental illnesses.

Nothing personal to anyone. ;)


lol its a fun read though. and some of the people actually know what they're talking about!

Why are there so many opinions and so few facts here?

What's certainly NOT an opinion is that, like most Sony baloney, it is designed first and foremost to look stellar on paper, to create the hardware as the perceived "standard". Any deviation from that = inferior, regardless of whether it's actually the case.

That, and of course the BR Trojan Horse.

your just using a theory that you thought up. the ps3 does not necassarly have more RAW power it only does that in a given situation. Also the 360 has a much more effective and powerful GPU and that is where a HUGE amount of what a game looks like . Epic openly admits that gears doesnt touch what the xbox can do and knowing what they know now they could do better.

I am not using a theory I thought up. It is well documented how powerful the cell is when compared to the 360’s cpu. And as you say "Cell is only more powerful in a given situation". Well, the cell is more powerful in every situation if programmed efficiently for it. A processor can only do so much without overclocking and liquid cooling. The software code is what mainly tells the tale and like a core 2 duo vs a Pentium D, the core 2 duo will always win if programmed efficiently. If not, then the Pentium D will surpass it even though it is inferior. I agree that the GPU of the 360 is more powerful than the PS3's. However the gap between the 360's and the PS3's GPU is not as wide as it is when comparing the CPU's of both systems. I would say they are neck and neck if anything. At the moment, the 360 has an advantage in that the game developers have had more time to work with the final devkit build for at least a year of the 360. Only time will tell if the wider gap the Cell has in power will be a factor sooner than later in allowing the PS3 to match and then surpass 360 game qualities which includes physics and more CPU intensive tasks in exclusive titles.

If the PS3 plans to catch up, it better hurry. The fist xbox was released in November 2001. The second was in November 2005. Therefore, the next could be around November 2009. That's only 32 months away.

Yeah, but was the ColecoVision processor more powerful that in the Atari 5200?

If the PS3 plans to catch up, it better hurry. The fist xbox was released in November 2001. The second was in November 2005. Therefore, the next could be around November 2009. That's only 32 months away.


The 360 will be around a few years longer than the original X-Box was. Unless game sales aren't doing as well for the 360 because of competition which I doubt will happen since the 360 is doing very well. The Ps3 may very well catch up and surpass the 360 in graphics/physics in upcoming games this year. However even if it does, having the best graphics doesn't mean Nintendo or 360 at that time won't be selling better. Right now the Wii is doing better than both the 360 and PS3 combined when you compare their launch numbers.. It will be interesting to see how things play out. All in all the competition between the systems is good and is bringing out better quality games much quicker. If Microsoft decides to replace the 360 as quickly as the original X-Box, I would view that as a defeat in a way, which leaves room for Sony to do the same thing it did this generation and release a more powerful system with a lot more hype. I would welcome another next-gen system mroe quickly since I am someone who always wants the best of the best technology as quickly as possible. With technology moving so fast, I can't imagine what the next-gen systems will be like..

The 360 will be around a few years longer than the original X-Box was. Unless game sales aren't doing as well for the 360 because of competition which I doubt will happen since the 360 is doing very well. The Ps3 may very well catch up and surpass the 360 in graphics/physics in upcoming games this year. However even if it does, having the best graphics doesn't mean Nintendo or 360 at that time won't be selling better. Right now the Wii is doing better than both the 360 and PS3 combined when you compare their launch numbers.. It will be interesting to see how things play out. All in all the competition between the systems is good and is bringing out better quality games much quicker. If Microsoft decides to replace the 360 as quickly as the original X-Box, I would view that as a defeat in a way, which leaves room for Sony to do the same thing it did this generation and release a more powerful system with a lot more hype. I would welcome another next-gen system mroe quickly since I am someone who always wants the best of the best technology as quickly as possible. With technology moving so fast, I can't imagine what the next-gen systems will be like..

There was supposed to be 5 years in between the PS2 and the PS3, but things got delayed a bit. The xbox waited 4 years inbetween upgrades.

4 years is really pretty long for a PC upgrade. The processors should be about 4 times as fast for the same money after 4 years. 6 years is definately too long.

"If Microsoft decides to replace the 360 as quickly as the original X-Box, I would view that as a defeat in a way, which leaves room for Sony to do the same thing it did this generation and release a more powerful system with a lot more hype"

That Sony come out second is considered by most a bad things. I bet MS would take that deal every year. They get the one year lead, Sony comes out second with some hype.

I didn't read through all 5 zillion posts of this thread but it is not necessary because only one thing matters...........


MOST games will look VERY SIMILAR on the 2 systems!!!!!!!!!!!!


They are both almost equally capable. On paper maybe it looks different, but what matters is the games, and they will be (and already are) nearly identical in multiplatform games. You would have to scrutinize the multiplatform games to be able to tell any difference. In the end, the television you are using will make so much more of a difference.

However, in other applications such as AI, physics, etc. there could be bigger differences. But again, it won't be night and day because they are both very capable. I think this issue is overhyped and blown way out of proportion. It is still very interesting though. I am glad I own both systems and I am enjoying GOW and RFOM right now.

Jedimastergrant,

Perhaps you do need to read the previous posts to understand why the games look similar.

But you're sort of right is about the games but for developing you have to do concessions to get the product out on multiple platforms.

There was supposed to be 5 years in between the PS2 and the PS3, but things got delayed a bit. The xbox waited 4 years inbetween upgrades.

4 years is really pretty long for a PC upgrade. The processors should be about 4 times as fast for the same money after 4 years. 6 years is definately too long.

"If Microsoft decides to replace the 360 as quickly as the original X-Box, I would view that as a defeat in a way, which leaves room for Sony to do the same thing it did this generation and release a more powerful system with a lot more hype"

That Sony come out second is considered by most a bad things. I bet MS would take that deal every year. They get the one year lead, Sony comes out second with some hype.


Remember that with continued support for the previous system as the Ps2 had and still has, most people can and would wait a year for a PS4 if Microsoft releases a newer console first. The PS3 had a bad launch because of unexpected shortages in Blu-Ray laser diodes..If they didn't have that shortage, it would have been considered a great launch. If the PS3 gains ground and becomes as popular as the PS2 was, they may be in a similar situation at the time Microsoft plans their future consoles release. I don't think there is any rush by Sony to come out with a console first this next time around.

I don't think there is any rush by Sony to come out with a console first this next time around.

Okay then, we'll have to agree to disagree.

Originally posted by Skogan
There was supposed to be 5 years in between the PS2 and the PS3, but things got delayed a bit. The xbox waited 4 years inbetween upgrades.

4 years is really pretty long for a PC upgrade. The processors should be about 4 times as fast for the same money after 4 years. 6 years is definately too long.
From a financial perspective, I wonder if its smart for MS to release a new console every 4 years, when normally you lose money on every console sold for the first year or so.

From a financial perspective, I wonder if its smart for MS to release a new console every 4 years, when normally you lose money on every console sold for the first year or so.


I don't think MS has a choice in the matter in order to stay in the game. I doubt that XBOX division will be profitable for at least next 5 more years (if they can break even). I am sure they're seeing this is as investment for the time being.

Brandy loyalty is a big thing to break (against Sony and Nintendo)

From a financial perspective, I wonder if its smart for MS to release a new console every 4 years, when normally you lose money on every console sold for the first year or so.

I'm not entirely sure Microsoft ever really expects to make money off their Xbox division. I truly believe the Xbox came about because Bill Gates was shopping for a console and said "I don't want a Sony PS2, I wish I had something better... if only I had billions of dollars to research and develop my own. Oh *(&t!, I do!!"

I'm not entirely sure Microsoft ever really expects to make money off their Xbox division. I truly believe the Xbox came about because Bill Gates was shopping for a console and said "I don't want a Sony PS2, I wish I had something better... if only I had billions of dollars to research and develop my own. Oh *(&t!, I do!!"

Bill Gates has his own virtual reality glider in his house which is the same one found in Epcot Center. I has many such devices worth millions. I doubt he is playing games on the 360 if anything..Both Sony and Microsoft are in the gaming industry for the money, it is making them millions even if they lose money on the console at first. It is also good advertising for the Microsoft products in general as well as for Sony since they all revolve around home entertainment in certain ways..

What i really want to know is, to what desktop CPU's the ps3 & x360 can compare.
Shoud a Quad Core QX6700 at 3,2 GHZ be faster/better then the Cell?
Is a FX60 faster/better then X360's Tripple CPU?

And my last question is, does a X1900XTX outperform the Xenon?


Thanks for answers

"It really is surprising to me that the PS3 had a year after the 360 launched and its hardware is inferior"

Actually its not inferior, they are simular, just work differently. In a nut shell the PS3 is much harder to develop for, and alot more expensive to develop for being that it takes alot more resources: time, people.

It seems as though hardware does not matter as much as it used to. We are at the mercy of game studios and developers, the highest end gear wont make a game look amazing if its not there to begin with.

Shoud a Quad Core QX6700 at 3,2 GHZ be faster/better then the Cell?
Is a FX60 faster/better then X360's Tripple CPU?

When it comes to games, a quad core CPU would blow the snot out of the PS3/360's CPU. Heck, a dual core, out-of-order exeuction CPU would probably be better.


And my last question is, does a X1900XTX outperform the Xenon?

That is difficult to answer. If you count the Xbox 360's eDRAM, then yes, the Xbox 360's GPU will probably outperform the X1900XTX/RSX.

When it comes to games, a quad core CPU would blow the snot out of the PS3/360's CPU. Heck, a dual core, out-of-order exeuction CPU would probably be better.



That is difficult to answer. If you count the Xbox 360's eDRAM, then yes, the Xbox 360's GPU will probably outperform the X1900XTX/RSX.

And if a pc was build with 2 X1900XT's in crossfire? would it be better then the xenon?

Thanks for ur answers

Seriously, all of you should stop this speculation. All of you are taking bits and pieces and forming arguments based on incorrrect logic.

Proof is in the games.

What i really want to know is, to what desktop CPU's the ps3 & x360 can compare.
Shoud a Quad Core QX6700 at 3,2 GHZ be faster/better then the Cell?
Is a FX60 faster/better then X360's Tripple CPU?

And my last question is, does a X1900XTX outperform the Xenon?


Thanks for answers

A PC and console are designed with different requirements and should be treated as such. They aren't the same thing.

Read the whole the thread and you can clearly see that.

You're making this into a PC vs console comparison. It's like comparing a Cessna vs an Extra 300. different goals in mind.

Since everyone else has been offering their opinions on this extremely interesting subject..

My backrgound is that of a programmer. Mostly mainframes, for about 20 years. About 10 years ago I got interested in how games worked, and filled my bookshelves with books on the subject. I learnt alot, namely how to program in C and Intel Assembler, and how to write a 3D graphics engine. BTW, what I learnt from that experience (apart from how useful Matrices are for 3D transformations!) is that a 3D graphics library is not so hard, but the maths required for applying physics to it was well beyond me.

I did actually write my own game (for the PC under DOS which I called Zaquinox) -a vertical scrolling shooter based on a similar game I loved playing for the commodore 64 called Uridium. I digress, but the game got me in a lot of trouble, because a fellow in Norway offered to "buy" the game off me and then talked me into trying to set up a game development company for him. Long story short I lost about 30 grand setting up the company only to find he had no money.

Anyway, with regards to this subject. The big thing I learnt about games programming, which the original poster never mentioned, is that you get to break all the rules of good application design.

Way back when programming PC games meant knowing how to squeeze every drop of performance out of the Intel processor, I read a book by a guy called Michael Abrash (he helped do optimizations for the Doom guys). I think it was called the Zen of game optimization.

It was nothing short of amazing the things you could do to save CPU cycles. From fundamental things such as partial or full "unrolling of the loop" (meaning you repeat the same instructions over and over again rather than having one or more instructions inside a loop) to subtle bit manipulation thus avoiding math instructions (especially divides!) eg. binary shifts.

Anyway, the big thing I got from understanding the principles of optimisation is that when you truly learn the CPU, you are free to do whatever the hell you want to get the results you are looking for. In fact you can even change the target to get those results. You can cheat, just as long as the user doesn't know they are being cheated, it's all good.

So I think that although it is possible and even reasonable to point out the PS3 is a pig to program for, (and I agree that it is) speculating what that means in terms of how far you can push games on it is probably not going to yield a lot of truth. We know from the past consoles that developers continue to get more from these machines over the entire life of the machine, as they look to trump the titles that came before.

Developers "can" cheat. I believe that in Oblivion (for example) they fundamentally changed the way trees are rendered to suit the limitations of the PS3. The end result is that the PS3 version actually looks nicer.

Of course, in the modern day world of multi-platform ports, it is very true that we may see games that don't take full advantage of the ported platform, but it's also true that game engines, optimized for their target platforms will emerge that will allow the best performance to be extracted from the target platforms without too much change to the higher-level game code.

Give them time (perhaps another couple of years) and they will "optimize" their ability to port games while taking more advantage of each platform in a more cost effective fashion.

My view is that in terms of the end results going forward, there will be very little difference between a 360 game and a PS3 one, but I do suspect that in 2-3 years time, games delivered to take advantage of the Blu Ray drive will create a point of difference that the 360 won't be able to emulate. 360 fan boys take note, I own two 360's and don't own a PS3 -this is just my honest opinon, and sure it could be wrong -I'm just speculating.

Comment removed

ps3isallhype
360guy

just by his/hers names he/she's biased and will prob never post again.

this thread must have been informative for Keyser to not have closed it :D :rolleyes:

ps3isallhype
360guy

just by his/hers names he/she's biased and will prob never post again.

this thread must have been informative for Keyser to not have closed it :D :rolleyes:

Try reporting it next time.

Kyser

This thread has gone on long enough...