Will conroe be a step back for ffdshow ?

Have you tried doing your resize to an even multiple of DVD resolution such as 1440x960 ?

My JVC SX21 projector is 1400x1050 and I do notice some lag/stutters when resizing to 1400x1050 vs 1440x960.

Resizing to 1920x1080 is a helluvalot of area for LimitedSharpen to work on.


But as far as Conroe goes, we'll have to wait and see once they start to ship.

I hope to be an early adopter of it. My current HTPC is Intel 640 OC'd to 4.09 GHz.


My settings are:

Import("C:\\Program Files\\AviSynth 2.5\\plugins\\limitedsharpenfaster.avs")

SetMTMode(2)

hqdn3d(2)

lanczosresize(1440,960)

MT("LimitedSharpenfaster(ss_x=1.0,ss_y=1.0,Smode=3,strength= 100, overshoot=7)")


FYI: In a previous thread here, it was shown ( via demo clips ) that Lanczos4resize is s waste of CPU with no visual benefit.


- Andy

mn1265, That's very helpful, I am as well looking for some benchmarks to get an impression how much more of performance can a Intel-machine give me really in comparison to my x2 3800 at 2.45ghz.


Andy for that purpose could you do me a favour and try to run my script at your machine and tell me what you see ?


Back to the subject, I just found an article ( in german) that stated:


"With advanced Digital Media Boost all 128-bit SSE-, SEE2- and SSE3-Instructions will be executed within one CPU-Cycle, which means practically a doubling of speed of these insrtuctions.


The out-of-order-Execution-pipeline has been drastically shortened from 31 stages to 14.


and:


Wide Dynamic Execution- 4 instructions can be finished per core within one cycle (instead of three so far)"


Reads all nice, but what does this mean for ffdshow ?

Quote:
Actually it's the other way around, the P3 architecture is much faster, clock-for-clock that the netburst architecture. Actually the P3 and K7 architectures were very close clock-for-clock, that's why the P4 was "owned" by the K7, especially in the early days before Intel got the clockspeeds of the P4 ramped sufficiently to make up the difference.


In fact if you look at the Centrino/Pentium-M architecture, it's heavilly based of the P3 architecture and similarly squashes Netburst clock-for-clock. My understanding is that Conroe is the result of Intel realizing the "mistake" it made with Netburst, mostly that people are smart enough to understand clockspeed isn't everything, but also that they were hitting the clockspeed wall, and the subsequent re-investment/development of more efficient (P-M/PIII/K7/K8) architectures.


P4's are known to do better than AMD chips at ffdshow, but that has very little to do with clockspeed (though that's how it's phased usually), and everything to do with ffdshow is just something P4'd do well. You can find numerous things that AMD chips do much better than P4s, despite the clock speed handicap.

Quote:
Whoa that is way off. stranger89 has it right. That is why the P3 got killed off so quickly - remember how good the Tualatins were performing? They were slaughtering the P4 Williamette, at lower clock speeds. It's ok if your competitor is beating you (sort of), it is horrific if your older cpu line at lower clock speeds is killing you. P4 didn't really take off until the Northwoods came out.


Just a guess on why ffdshow runs better on P4's, it has to do with the architecture. In this case MHz or clock speed does matter. The netburst architecture has a very long branch prediction/recovery pipeline, this is the feature that allows the cpu to ramp up in frequency (I think because each pipeline stage is less complex). Prescott has a 31 stage pipeline. From an article at techreport, this pipeline operates in a speculative nature, a branch prediction unit predicts what's going to happen next. Great if it's right, bad if it is wrong since the pipeline is so long it has to retrace further.


Now, how does this affect video? Comparitively speaking, video applications have done well on the Netburst architecture because the instructions are fairly predictable. Games and other math applications are more unpredictable, hence the P4 tends to do worse, since whenever the branch predictor is wrong, it takes longer to recover. That's why cripped Celerons on the P4 usually hold their own when it comes to various video applications. And this is where the higher MHz actually does help.


So what will we see with Conroe? Well it appears that the IPC has actually been increased (I've seen 4 IPC quoted) to make up for the somewhat lower MHz, and there are other improvements. All indications point toward a kick ass cpu, and I will be first in line to own one as well.