Mark_A_W, that's what I thought (about the lenses). And I know that my 8" probably won't give a good image at 1600x1200, but I'm willing to try anything once (well, almost anything...) to get good PQ. What resolution would you recommend for a Barco Graphics 808 for use with ffdshow? No other considerations, only the best image possible with ffdshow. I hardly even see the Windows desktop since the change into "theater mode" is automated by Girder. I use PowerStrip.

gazzagazza,

I noticed that also.

Using overlay and elecard bulid 2510 this is the colourspaces used at each step (info provided by zoomplayer) (PAL DTV file playback)

There is no reference to colour space until the MPEG2 Spliter is hit.

MPEG Splitter output RGB 0 bits
Elecard Video input RGB 0 bits
Elecard Video output YUY2 16 bits
FFDShow input YUY2 16 bits
FFDShow output YUY2 16 bits (this can be changed and is 10bits with
YV12 in use)
Overlay input YUY2 16 bits
Overlay output YUY2 8 bits
Video Renderer Input YUY2 8bits

So it seems YUY2 is perhaps the best to use with the least colour conversions?
I tested a DVD with WinDVD 5 codec also and it also uses YUY2.

Note this testing was with PAL source, so not sure what happens with NTSC.

Im curious why it is 16bit the entire way but only 8bit out from the overlay mixer/render thingy?

Graphics card is 9600 Pro. (testing done on DVI head with VGA adaptor to monitor usually use VGA head to projector).

Gazzagazza

If i read you post correctly you are saying your system is not even coping with an FFDSHow resolution of 960x540 which is also your desktop res. If this is so i don't think it is your video card. I am running higher than that with a 9200SE which is much slower than the 9000 Pro. Moreover the only the scaling the card is doing is from the FFDShow output to the desktop resolution - which in this case is zero.

I cannot remember the resolutions i ran my 9000 Pro at but it was much higher than those you are having problems with (Pentium 2.4G). I think i recall running 1440*1152 with Lancoz (sp?) - at 50Hz mind you.

Bill

Sorry for "anamorphic" confusion. That happens when digital guy messes with CRT.

Mark,

you will be in for a long wait. It will probably be 2 years when HD2+ will hit the price point you are mentioning. Anyway, digital is just easy. No need to recalibrate, wonder what is the sweetspot. Just sit back and enjoy the show with all that digital punch that you do have to "grow into".

As far as how much can 9000 Pro take in terms of input resolution I don't know, but I know that other parts of Gazzagazza's system are capable of much more than he is getting. Video card gets more and more busy as you increase FFDShow output resolution, that is the fact. If you use overlay that is probably not an issue, but for VMR9 it definitely is.

Spoonfed,

I think you've hit it right on the head. There was a lot of discussion how ATI's overlay is superior to Nvidia as it uses 10-bit output, as opposed to Nvidia's 8-bit overlay. Looks like overlay is overlay and 8-bits is just hardware limitation. That is the reason overlay color rendering does not look as rich as VMR9. In VMR9 9600 and 9800 series cards (and Nvidia FX 5200 and higher) render color with 128 bit floating point precission and then output signal without dropping the quality to 8-bits at any point.

The fact that YV12 has only 10-bit depth does not mean it carries signal of "lesser" quality than YUY2. YV12 is 4:2:0 (compressed chroma) and YUY2 is 4:2:2 signal (expanded chroma). While current decoders output YUY2, I Forceware Multimedia and TT 2.0 will both output YV12 to FFDShow. At the end of the day this causes an extra colorspace conversion, but helps speed up FFDShow processing and assures proper hardware chroma upsampling.

It would be great if you could check VMR9 graph. I hope that only difference would be YUY2 16-bit as opposed to YUY2 8-bit.

My concerns about the ffdshow "Levels" adjusters were misplaced - they seem to work fine. I tested output for YV12 scaled to output 0mv to 714mv and with YUY2 output scaled down to 44mv for black and 655mv for white. Both ways it worked fine.

Of course it should be remembered that digital gain scaling is lossy (unless it is by an integer multiple and not truncated) - one level in seven is lost when scaling to 0 to 714mv so it is best not to do this at all or only once if necessary. Gain reduction is also losssy but I don't know the details. I can't see any difference in projected greyscales with gain scaling in my system. Hopefully the digital signal chain is not being gain scaled anywhere else. Coders beware though!

I presently am able to use Spline resize to 1440x960 with about .5 sharpening with no other filters and use YUY2 output. My frame rate is about 23.9 and there are only ocasional tiny judders. This has almost eliminated the "it's alive" look of large surfaces for me and there is a lot of potential sharpening possible if wanted. If needed, using YV12 with Levels scaling should be very similar with less CPU demands.

Charlie

Hi guys,
How does RGB32 output from ffdshow compare to YUY2 and YV12? I have absolutely no idea what you guys are talking about regarding colorspaces. :(

Hi guys,

I have several problemes that maybe you know how to solve ;)

I'm using a Dering + Gradual Denoise + Unsharp Mask on 1280*720p material via ZP. The image is really good, but it's not as smooth as without using ffdshow. It seems like the ffdshow-20021213.exe version is bit smoother than the ffdshow-20040610c_SSE2.exe version. In fact, I get some drops on slow pans or on very fast moving scenes.

FYI, I have P4C 3.2 with 512Mo and a Radeon 9700 Pro.

The other problem is that whenever I go in ffdshow configuration, my overlay settings are lost (contrast, saturation, hue, etc...). Is there anyway to fix this problem (I have it with both versions of ffdshow) ?

Last one, it seems that ffdshow is not really compatible with Hyper-Threading. When I try a 1920*1080 resize, my cpu0 is doing nothing while my cpu1 is at 100%, therefore it gets really choppy :(
From what I've been told, there is no problem if the Hyper-Threading is disactivated...

Thx in advance ;)

vpopovic,

8bit does seem a limitation. Don't DVD Players (STB) output 10bit?

I will try VMR9 tonight. Am i correct here in saying Direct3D performance settings will affect CPU usage with VMR9?

Will let you know what the 9600 outputs then.

N3W813,

I tried RGB32 output but found some artifacts were introduced on certain colours.

Originally posted by Bill Green
Gazzagazza

If i read you post correctly you are saying your system is not even coping with an FFDSHow resolution of 960x540 which is also your desktop res. If this is so i don't think it is your video card. I am running higher than that with a 9200SE which is much slower than the 9000 Pro. Moreover the only the scaling the card is doing is from the FFDShow output to the desktop resolution - which in this case is zero.

I cannot remember the resolutions i ran my 9000 Pro at but it was much higher than those you are having problems with (Pentium 2.4G). I think i recall running 1440*1152 with Lancoz (sp?) - at 50Hz mind you.

Bill

Hi Bill,

I can do 960x540 desktop OK using bicubic, since MB RAM and CPU upgrade I can do 1440 x 540 bicubic. Anything more and there is trouble. Try 2 x DVD res and it really chokes. The only thing common between the two configurations I've run in the past year is the graphics card... so I will try a change to a 9600...

Originally posted by Spoonfed

8bit does seem a limitation. Don't DVD Players (STB) output 10bit?



Data is stored on DVDs in YV12 format. This means that there is only 8 bits of infomation for each sample.

Where the extra bits of precision comes into play is in the decoding process. There is math involved and if you only carry 8 bits of info you get rounding errors which ultimately manifest themselves as noise, bad colours, etc.

Most DVDs players now use 10 or 12 bit DACs on the output. This allows them to offer better resolution for "in between" values that are created when they oversample the output.

Note: for info on the colourspaces you can check out fourcc.org.

N3W813,
How does RGB32 output from ffdshow compare to YUY2 and YV12? I have absolutely no idea what you guys are talking about regarding colorspaces.RGB32 has video level 16 mapped to 0mv and video level 235 mapped to 714mv.

Your confusion with colorspaces is understandable since we are mostly involved with one digital electronic representation of colorspace Y'CBCR (read that Y prime C sub B C sub R) which has a number of ways of being transcoded into lower quality video streams. Each coding scheme both reduces bandwidth and specifies how the data is stored. There are literally dozens of coding schemes and each has a name. In essence all lossy coding schemes keep all or as much luminance data as possible and selectively reduce chroma bandwidth (resolution) with filters or averaging.

The reason why analog R'G'B' (fresh from the camera gamma circuit) is not used directly is that it costs too much for all the circuitry and is also hard to do things like filtering without a large amount of additional expense. All of the digital coding schemes used in consumer gear are lossy in that they reduce the data rate to a level that can be handled by cheap electronics by throwing out (valuable?) data. None of these schemes are desirable, as they are lossy, but some are less desirable than others depending on your application. A few coding schemes are so close in application that is is debatable which is better and the debate goes on...

Charlie

A quick trip to www.fourCC.org shows the formats of all normal video formats. Here they are:

RGB 24 (or 32) 8 bits for each color, 24 bits per pixel.

YUY2 16 bits per pixel. Y Horizontal Sample Period = 1, V & U Horizontal Sample Period = 2. Y & U & V Verticle Sample Period = 1.

YV12 12 bits per pixel. Y Horizontal Sample Period = 1, V & U Horizontal Sample Period = 2. Y Verticle Sample Period = 1, U & V Verticle Sample Period = 2.

YUY2 and YV12 are similar functionally, YV12 has half the vertical chroma resolution and different storage formats. There are probably other differences as well.

If DVD is really YV12 then it seems a shame to convert it to YUY2 since the chroma data has allready been tossed. There's probably more to this transcoding than meets the eye though.

Charlie

Originally posted by Charles Black
If DVD is really YV12 then it seems a shame to convert it to YUY2 since the chroma data has allready been tossed. There's probably more to this transcoding than meets the eye though.

Charlie

It's a shame, but unfortunately necessary. This extra colorspace conversion is less costly if done with the same (similar?) algorithm used to upsample chroma in YUY2. For example, Forceware multimedia will convert YUY2 back to YV12 using its own algorithm, rather than FFDShow algorithm. This hopefully reduces the loss as YV12 is "reverse engineered" using the same (hopefuly) algorithm. The benefit of the YV12 is that you get hardware chroma upsampling (probably YV12 to RGB32) that should be better than software. This depends on the codec, but you can test it easily in DVE chroma resolution pattern. Most common error is that color bars on the left of the screen are not the same color as the ones on the right of the screen, while they should be.

Spoonfed,

I am not sure whether 3D settings will encumber CPU more, or is it VMR9 itself that is consuming resources. The fact remains, CPU usage with VMR9 is more than with overlay. 3D settings will (or at least should) impact GPU usage where AA and AF might cause tearing and even stuttering. LOD detail for textures should also affect the image, and keeping it at highest image quality is probably the best setting. You can also try adjusting the LOD detail in registry to a negative value (i.e. -3), but I am not sure if ATI drivers will take that. Nvidia still takes -15/+15 range through registry while only 0/+3 are available through driver settings. I believe WHQL test still requires -3/+3 levels for the drivers. Negative settings produce arguably sharper image, in a strange way, as textures are not rendered so smooth. I used to like it, but now I believe 0 is the best setting.

Vlad,

A question. Why would the nVidia mpeg2 decoder not just stay with YV12 and not use YUY2 at all?

Charlie

short info:

bugfixed version is up, for the banding and chroma bug reported by pmd918 and TheLion

Milan also added a realy nice new feature, u can now auto resize by a multipler.

PS: no new resizer im still working on it

Theres also colour space discussion in the WINDVD6 thread (i think)

Mention of a registry entry for WinDVD to process in YV12.

However im not sure if this is right, as I mentioned earlier it is YUY2 16bit before the codec (but this may well be as the codec has requested this input?)

As for the 8bit overlay to renderer output. Does that mean that a STB DVD player is better than a PC can manage? OR are DVD STB's only 10 or 12bit internal processing and 8bit like PC on the last rendering stage?

Originally posted by Spoonfed
As for the 8bit overlay to renderer output. Does that mean that a STB DVD player is better than a PC can manage? OR are DVD STB's only 10 or 12bit internal processing and 8bit like PC on the last rendering stage?

I think the answer is the infamous "it depends".

All of the newer high end cards from ATI and Nvidia have 10 bit DACs so they always end up outputting at 10 bits. It's unclear to me how they do the 8 to 10 bit mapping.

Equally important is the DAC frequency since this relates to the oversampling capabilities of the output. ATI runs theirs at 400 MHz, not sure about Nvidia. The Denon 5900, as an example, has 12 bit DACs but only clocks them at 216 MHz.

Which is better? I'm not sure but I don't know if anyone has done a head to head using the component outputs.

Does a 9600pro 128 count as "newer"? ie 10bit DAC's? I know its 400mhz.

Im keen to try VMR9 now and see if colours improve. 8bit just seems hmmm wrong to me :)

Charles Black,

I am not sure. As I mentioned, YV12 is faster to process in FFDShow and at the end of the day use of hardware chroma upsampling seems better than using software (in this case arguably chroma is downsampled using the same upsampling algorithm for minimal loss). Overall, I agree with the aproach. Very good results with speed increase. I did not go into analyzing if YUY2 all the way would be better. Obviously they are not able (at least at this stage) do decode MPEG2 stream directly into YV12.

Spoonfed,

You got your answer as far as DVD players vs. HTPC. Just to mention, there is a famous MP-1 mode that alters the DAC section of the video card output and is meant for hi-end analog market (i.e. driving reference level CRT projectors). People say it does work well. At one occasion I had the opportunity to try my HTPC agains Lexicon DVD player (both through analog outputs, Infocus 7200, 96" screen) and I must say that HTPC was better. This was with Quadro video card that (if nothing else) is known for very well set analog stage.

9600 Pro is definitely one of the "new cards". In VMR9 it also has the ability to process video with 128-bit color precision. You are probably in for a surprise how little difference between 8-bit and 128-bit processing there is. IMO VMR9 is definitely better, but you need to pay attention to notice it, unless you are a die hard videophile.

gazzagazza,

Does your video cable have pins in both connectors? In other words can you swap ends with it?

If so then you can use two D to BNC cables with 5 BNC "T's" to get at the signals. This would be good since it is non-destructive and easy to keep noise from getting into scope. These cables are ubiquitous so maybe you can do this for free.

The scope you have should be easier to use than the ones I use since it has some sort of video trigger that will allow you to pick which line to look at.

Charlie

AndyIEG

Thanks - downloading for trial tonight.

Bill

14 answers since my questions, not even one for me :)

Just starting to play around with Overlay vs. VMR... is there something special I need to do to enable VMR in Zoomplayer?

BangoO,

Try turning hypertrhreading off. That should give you a little more CPU time.

Try less sharpening and filtering. I can't use as much as others are using without getting a lot of artifacts and noise. I don't use anything except spline resize with .5 luma sharpening. I expect that more sharpening would be Ok but need to check it out slowly.

Charlie

Madpoet,

Open up Player Options in ZP and goto DVD/DVD Setup and select VMR9 Windowed. Then goto Filter Control/Settings and select Fix VMR9 Scaling Bug.

Charlie

Am I right in saying that there is going to be SSE2 optimization for resize as well? I am using the latest SSE2 build from today and I must say I am impressed that my CPU can hang in there. I can do gradual denoice at 40 then resize to 1776x1000 (my desktop resolution when playing back movies) with luma sharpening at 1.6 or so. This results in pretty high CPU useage with DVD but definately useable. For xvid, divx, mpeg movie files (basically the 700MB full-length movies..some bigger), I can use denoise3d at 0.5,1.0,5.00 HQ in front of the same resize settings with no CPU maxing out. Of course, this is with overlay...not VMR9 on my P4 2.4Ghz (533FSB). I hear that I may actually get better color reproduction (maybe performance) if I switch my Radeon 9600SE with this Quadro FX 500 I have. Anyone have an opinion on that? Can I expect to squeeze more performance from builds in the near future? Maybe enough to get denoise3d to work on DVD as it does with video files?

If one had to choose between resize and denoise3d, which would you choose? I'm kinda in that boat with DVD playback and I don't see me upgrading the HTPC anytime soon (would have to replace the motherboard, CPU, and memory and WOULD trade it with my personal system but my HTPC case is uATX not full ATX).

Also, is anyone using powerstrip with the Quadro card (HDTV via transcoder not DVI)? If so, is it safe to say I will be able to get the same or similar resolution results in powerstrip that I do with my current Radeon card? I only like to use 640x480p and 1776x1000i right now...but I like them both very much. Thanks.

BangoO,

I think part of the reason that you didn't get much response is that the filters you are using are not the most common. These days, the filters that most are using, in order of preference, are:

Denoise - using denoise3D
Resize/sharpen - using Lanczos
Levels

I have a very similar setup to you (P4 3.06GHz, 512MB PC2100, Radeon 9200) and I am able to run all of those filters plus the Properties filter, and resize to 1440x960. I have my cpu overclocked to 3.4GHz.

On your hyperthreading question, Charles' suggestion is correct, turn it off. I did a test on my system and there is a slight but measurable improvement in performance when HT is turned off.

Dan Bennett,

Yes, Andy is working on an optimized resize routine. Owen, one of the recognized ffdshow experts, says to run denoise3d in preference to resize/sharpen if you don't have the power to do both. Other experts in this thread recommend the opposite. So you probably should try both and see which YOU prefer. Don't know about your other questions.

Hope this helps,
Phil

Maybe I'll be able to run both like I can for video files! All I can do is wait and cross my fingers...

Does anyone have any thought about using a Intel Pentium M 90nm 2.0GHz with the 2MB L2 cache (this is not a prescott, it's the new "mobil" cpu with 478 pins and can be used on a desktop) ??? It should run fairly quick I think ????

I was up till 5AM last night fiddling around trying to get ffdshow and Zoom Player to work on my Sammy HLN4365W. Looks great. I have it going through my VGA cable still (might see if I can get the DVI output to work, although my TV has 1 DVI input, so it would be just for kicks, unless it's amazing enough to warrant getting another DVI cable and switching them between cable box and PC), ffdshow resized to 1920x1080, Blur & NR and Unsharpmask on (lower than suggested settings on first post of this thread - was too sharp and produced jaggies and looked like lower resolution), and a few picture adjustmestments for brightness/gamma/luminance.

It looks fine right now, but I was wondering about dscalers. Someone said that going from 1080 with no dscaler, to 540 with a dscaler, he was able to get a much better PQ. Can anyone enlighten me about dscalers and how to set them up and tweak them?

BTW, my computer is an HP Media Center, P4 2.8GHz with 512 RAM and an NVidia GeForce 5200 Ultra 128 RAM video card (using VGA-out, also has DVI). I have the NVDVD and InterVideo(?) programs on my computer, and am using the InterVideo video codec in Zoom Player.

This is in comparison to my Panasonic S25 DVD player, which I bought modded from a Japanese bookstore chain to watch Region 2 Japanese DVDs as well.


BTW, besides the test patterns and everything, what do you use to calibrate your picture? What movies have good sharpness/focus? What about lights/darks?

I've found the spider cave in Harry Potter & The Chamber of Secrets, the close-up of large spider's face, to be great for fine-tuning brightness (meaning very small, single-digit increments). I used it last night to tweak ffdshow settings (gamma/luminance/brightness, etc.) to get it just right. I'm trying to leave my TV's settings alone, as all have roughly the same settings, and when I watch a movie/TV/game during the day, I just turn up the Brightness from 52 to 70-75 to compensate for the amount of light our living room gets - then turn it back down to watch at night.

A week or two ago I noticed The Rundown had a really great picture - surprisingly sharp (on my DVD player even). Unfortunately this was a rental, so I can't check it on my new PC DVD setup (although I am on the Blockbuster DVD Freedom Pass till next week, so I could check it out...). LOTR-ROTK was kind of a letdown in the PQ department, and I'm hoping the extended version, spread over 2 discs with all the extras on two other discs, will have a higher sampling rate for better quality. But I'll have to try watching it again on my PC now to see the difference.

Any other suggestions for settings/adjustment, or just plain to be "wowed" by the great picture? Or any (popular) ones to try not to watch for PQ evaluation? (Star Wars Ep 1 wasn't very sharp at all, at least on my regular DVD player).

I.e., what DVD to use as a baseline for my ffdshow/ZP tweaks?


I tried doing some searches on this thread last night, but with 75+ pages of 2+ years' worth of posts, it's a lot to wade through. Any help would be appreciated.

One more thing.

If you tweak with a softer-focus DVD (i.e., SW-TPM or LOTR-ROTK) and get your settings "set," then watch one that's much sharper (i.e., The Rundown, Gladiator), does it look overly-sharp and get jaggies?

I imagine it'll take testing it on a number of DVDs to average out the results.

But with something like DVE, can I use that to do the overall settings? I've noticed that at least for my TV's sharpness settings, I can't really notice any difference between 0 and 100. So will fine-fine-fine tuning the sharpness, etc. using DVE on my PC help a lot, so I don't have to tweak and tweak for each DVD I'm watching? Is that how everyone else does it?


Thanks.

FFDShow supports a number of profiles or image settings. You can use these to setup for media that requires somewhat different processing (like your sharp and soft DVDs). Then use the configuration app to choose which you want to use before launching the player. I believe that this appears in an 'Image Settings' tab near the top of the list.

There may also be hacks to allow you to set the image settings profile automatically, but I have not explored that.

At least you don't have to remember which tweaks to use for each media type or quality.

Originally posted by cyberbri I.e., what DVD to use as a baseline for my ffdshow/ZP tweaks?
Start by calibrating with Avia and/or DVE. Then check out the Reference DVDs (http://www.avsforum.com/avs-vb/showthread.php?s=&threadid=372989) sticky thread. It's best to use a film you already know.
Originally posted by cyberbri
If you tweak with a softer-focus DVD (i.e., SW-TPM or LOTR-ROTK) and get your settings "set," then watch one that's much sharper (i.e., The Rundown, Gladiator), does it look overly-sharp and get jaggies?
Yes, DVDs with tons of edge enhancement (e.g., Gladiator, Terminator 2) will look worse with sharpening because it accentuates the halos around contrasting objects. You can set up multiple profiles in ffdshow, and it should be possible to have it key in to the media name or serial number to automatically activate a "soft" profile for certain DVDs. Alternatively, you can turn on the keyboard control and turn off the resize/sharpen when watch a problem DVD.

Originally posted by pmd918
I have a very similar setup to you (P4 3.06GHz, 512MB PC2100, Radeon 9200) and I am able to run all of those filters plus the Properties filter, and resize to 1440x960. I have my cpu overclocked to 3.4GHz.
On HD 720p material ?
I doubt it...

Thx for the answer anyway, I made lots of tests and the only advantage I can see on the resize over the unsharp mask is a little bit less of EE. I'm not testing anything on DVD material which is a lot less clean than HD material, so maybe that's an explanation.

About turning HT off, I'll try but will there be a "HT compliant" ffdshow version one day ?

BangoO,

Sorry, I overlooked the fact that you were talking about HD source material. Probably another reason why you didn't get much of a response. Most people here are trying to make DVDs look good.

Phil

I know pmd918 :)
I thought maybe someone would have an idea why It seems to loose some frames even my cpu1 is never at 100% (I'm not using resize).

Originally posted by BangoO
I know pmd918 :)
I thought maybe someone would have an idea why It seems to loose some frames even my cpu1 is never at 100% (I'm not using resize).

What ffdshow versions did u try? Does the frames always drop with all versions? Did u try overlay vs VMR7/9?


PS: question will a ffdshow version with just lanczos and spline resize work for u dudes or is there realy a importand reason to keep the other modes?

I'm trying to get denoise + resize working. My P4-2.4GHz can manage Gradual denoise = 22, followed by Resize at the screen resolution (1280x1024) with sharpening set at 1.3/1.3. However, I cannot switch from Gradual Denoise to Denoise 3d (.5/.5/5.0/HQ) without having major judder problems. Is there a significant benefit of one versus the other, from a quality perspective?

I'm also considering overclocking this thing, but I'm a bit worried about overheating it, and I don't want to add a bunch of extra fans, due to the noise factor. Currently I've got a Zalman copper heat-sync with its rather quiet fan, the Enermax supply with its two fans, and the ATI AIW 9700pro with its little spinner, and the system doesn't make too much noise. I used to have an additional two fans (supposedly quiet ones) and it was making far too much noise.

Perhaps when the new ffdshow with optimized resize comes out, I'll be able to do the Denoise 3d instead of Gradual Denoise. However, the new TheaterTek 2.0 is coming out soon, and it has the option of audio-decoding within the PC. I would like to try this, but then I'll probably run into further processing power shortages. :(

-=> Mike Hanson <=-

Andy,

You are always going to run into people who want something else. My guess is that most people currently use Lanczos because it gives nice PQ at cpu loads that are manageable. Since you have to start somewhere, IMHO that's the logical place. Spline is of interest to me because it should perform better than Lanczos.

boxsoft,

With your cpu, it's asking alot to manage denoise3d and resize/sharpen to such high resolution. Many people here agree that denoise3d is significantly better than gradual denoise. In fact, some people think that denoise3d is more important than resize/sharpen, so you might try running denoise3d alone to see how it compares to you current configuration. You might also try a combo where you don't resize to such high resolution (e.g. 1280x720). In any case, it is recommended to set Chroma sharpening to 0, since it seems to do nothing other than introduce noise.

Hope this helps,
Phil

Originally posted by AndyIEG
What ffdshow versions did u try? Does the frames always drop with all versions? Did u try overlay vs VMR7/9?
As I said in my previous post, ffdshow-20021213.exe and ffdshow-20040610c_SSE2.exe, the 1st seems to drop less frames.
I don't want to use anything else than overlay.

Originally posted by Charles Black
gazzagazza,

Does your video cable have pins in both connectors? In other words can you swap ends with it?

If so then you can use two D to BNC cables with 5 BNC "T's" to get at the signals. This would be good since it is non-destructive and easy to keep noise from getting into scope. These cables are ubiquitous so maybe you can do this for free.

The scope you have should be easier to use than the ones I use since it has some sort of video trigger that will allow you to pick which line to look at.

Charlie

Thanks... I've been looking at my options on this. As the 'scope doesn't load the signal, is it OK to use a VGA Y adapter with one side of the split going to the video display and the other side of the split going to a VGA to BNC cable (DB15 to 5 BNC). In effect this is a VGA passthru with the bncs enabling the scope to look at the individual lines....

Originally posted by BangoO
I know pmd918 :)
I thought maybe someone would have an idea why It seems to loose some frames even my cpu1 is never at 100% (I'm not using resize).

I don't know why are you loosing frames. Most common reasons would be overloaded CPU or GPU. Are you measuring your CPU utilization with full screen video or small window? What video card do you have? What are you temperatures for GPU and CPU under full load?

I know though that resizing (or doing any other processing) original 720p WM9 clips (downloaded from MS website) only worsens the image in my setup. I get absolutely smooth playback with resize of 720p clips to 1920x1080, but that seems really redundant for playback on my 720p display. I did not play with downsizing 1080p clips to 720p discouraged by these results and obviously lack of real life 1080p material. I can see there might be benefit to resize and post-process broadcast HD, but can't help you there - I am stuck with cable in NYC.

Originally posted by AndyIEG

PS: question will a ffdshow version with just lanczos and spline resize work for u dudes or is there realy a importand reason to keep the other modes?

When you are kind enough to ask, I think it would be nice to have bicubic. That is kind of basic, no-ringing algorithm that works well on its own as well as in multiple resize scenarios like Avisynth Lanczos4 and then FFDshow Bicubic.

I second vpopovic’s request for optimized Bicubic resize.

I have been doing a lot of testing lately with the aim of removing (as much as possible) all artifacts and sparkling high lights that contribute to the digital video look, and limit the overall performance of DVD playback.
To that end, I have actually come to prefer Bicubic over Lanczos resize. (shock, horror, amazement) :D

Sinc or spline may over time prove to be better, but at the moment they are to slow for serious resize use.

On my display, extra sharpening is not required so I don’t need to sharpen in resize.
I find that Lanczos can look a little harsh and tends to emphasize noise and sparkling high lights, much in the same way as Overlay rendering.
Bicubic resize produces a smoother and cleaner image with a less digital look, just as VMR9 does compared to overlay.
With the high resize settings that I use, overall sharpness and detail is about the same between FFDShows Lanczos and Bicubic resize algorithms.

If extra sharpness is required, resize Luma sharpening can always be applied.

I should also say that VMR9 used without resize or at low resize settings produces a very soft image for me. Bicubic resize may also produce a softer image at low resize settings, I did not test for this.
So VMR9 or Bicubic resize may not be the best option for displays requiring extra sharpening or for systems that do not have the power for high resize settings.

Remember, the sharpest image is not always the best image.
My aim is to produce the best image clarity combined with the lowest possible digital artifacts and noise.
It is possible to have both.
So fare, I have found the best results to come from using:

Denoise3d L0.5, C1.0, T5.0 HQ

Very high Bicubic resize (minimum of 2x DVD resolution but preferably much higher with no Chroma sharpening)

VMR9 output rendering.


As a side note, I have seen people recommend disabling Hyperthreading on there P4 systems.
I have tried this on several occasions and found that disabling Hyperthreading on my system does NOT improve video playback performance with the system pushed to the limit, but it sure does degrade system responsiveness to user input and general system performance.
Remember that the system is ALWAYS doing multiple tasks when playing video. Just reading the hard disk or DVD is a separate task.
Hyperthreading is a great P4 feature and should not be disabled IMHO.
Systems with Hyperthreading are noticeably more responsive and faster the non HT systems.

It does not hurt to try disabling HT on your system, just to see if it helps, but don’t assume it will be faster.

Regards,

Owen

____________________________
The FFDShow resize-sharpen dude.

Gazzagazza,

The Y cable adapter will work fine but you may notice some slight degradation in your video due to reflections from the sample ports. This is not a big deal and may not even be visible. The shorter the better on the Y. The shortest isthe BNC "T" which only is about 1cm long. When the sample ports are this short the reflections are still in phase with the original signal.

If you get two D15 to BNC adapters you can use BNC cables to stretch if necessary and have short sample ports. I bet this is a non issue though. I have a scope in my monitor R signal all the time and since I go From 15 pin D to 5 BNCs it makes sence to have the "t"s at the monitor. My scope requires triggering on the V and on the H lines as well as the video so I have 75 ohm BNC cables for the V and the H that are terminated high impedance and use a 10 Megohm scope probe right at the video "t". Many probes have a special adapter for sampling BNC male and a similar one for the female. They reduce the noise very significantly and are recommended.

If you care about the accuracy of your measurements you will have to compensate your probes. It is easy to mess this up on my scope since the shorter calibrator pulses that my scope uses for higher sweep rates always show a relatively flat topped square wave. In the old days they always used a low frequency square wave and it was hard to screw up. If in doubt seek someone with a bit of experience and they can show you how it's done in a few minutes. If you have to go it on your own I can write out the steps for you.

Charlie

Charles Black,
Just as a matter of interest, what bandwidth scope is required for effective measurement?
I wanted to check for ringing on my signal cables.
Is a 60Mhz scope adequate?


Regards,

Owen

Owen,

That will do a good job - much faster than we used to use. I don't think that a BW of less than 10Mhz will be very good for video. I have an old Tek 323 that I think is real slow that I'l put on the video tomorrow to see what it looks like. If your scope has delaying sweep and calibrated vertical amp you can do more than just ringing. All you need is a good probe. If it is really simple and only has a single sweep you can sometimes use tricks to delay your trigger a bit too. What kind is it?

Charlie

Thanks, will play once I get a project I'm in the middle of sorted. I take the point on reflections... I don't intend leaving it inline, I will take the PC to the workbench to do the measurements etc... Plenty of experience with instruments, just haven't delved into video signals before...

Originally posted by Charles Black
Gazzagazza,

The Y cable adapter will work fine but you may notice some slight degradation in your video due to reflections from the sample ports. This is not a big deal and may not even be visible. The shorter the better on the Y. The shortest isthe BNC "T" which only is about 1cm long. When the sample ports are this short the reflections are still in phase with the original signal.

If you get two D15 to BNC adapters you can use BNC cables to stretch if necessary and have short sample ports. I bet this is a non issue though. I have a scope in my monitor R signal all the time and since I go From 15 pin D to 5 BNCs it makes sence to have the "t"s at the monitor. My scope requires triggering on the V and on the H lines as well as the video so I have 75 ohm BNC cables for the V and the H that are terminated high impedance and use a 10 Megohm scope probe right at the video "t". Many probes have a special adapter for sampling BNC male and a similar one for the female. They reduce the noise very significantly and are recommended.

If you care about the accuracy of your measurements you will have to compensate your probes. It is easy to mess this up on my scope since the shorter calibrator pulses that my scope uses for higher sweep rates always show a relatively flat topped square wave. In the old days they always used a low frequency square wave and it was hard to screw up. If in doubt seek someone with a bit of experience and they can show you how it's done in a few minutes. If you have to go it on your own I can write out the steps for you.

Charlie

As a side note, I have seen people recommend disabling Hyperthreading on there P4 systems.
I have tried this on several occasions and found that disabling Hyperthreading on my system does NOT improve video playback performance with the system pushed to the limit, but it sure does degrade system responsiveness to user input and general system performance.
Remember that the system is ALWAYS doing multiple tasks when playing video. Just reading the hard disk or DVD is a separate task.
Hyperthreading is a great P4 feature and should not be disabled IMHO.
Systems with Hyperthreading are noticeably more responsive and faster the non HT systems.

It does not hurt to try disabling HT on your system, just to see if it helps, but don’t assume it will be faster.
.

Owen,

in my opinion disabling Hyperthreading in the BIOS is only good für debugging problems related to applications or drivers that are not multiprocessor/hyperthreading safe. It will probably not show better performance because the underlying HAL that Windows XP uses is still the Multiprocessor HAL. In fact, your experience that Windows became sluggish after disabling HT is more logical.

By itself the Multiprocessor HAL is slower than the single processor HAL´s that are available because there is some overhead involved in distributing threads to different processors. If more processors are used, or hyperthreading is active this will be compensated for. But if one uses the Multiprocessor HAL with one active processor ( which you would do if you disable Hyperthreading in the BIOS ) the system is slower as it could be.

So, the solution would be to not only disable Hyperthreading but also to change the HAL. This can be done by changing ACPI Multiprocessor HAL in the device manager to ACPI PC or ACPI Single processor. HOWEVER, in my experience, doing this can often cause a lot of problems because at the next boot Windows will try to redetect all hardware again.

The only really safe way to do is would be to reinstall Windows completely from scatch with Hyperthreading off.

This will trigger the Single Processor HAL to be installed and with this HAL Windows should perform better with 1 processor active.

Reactivating Hyperthreading would then trigger the Multiprocessor HAL to be installed so before doing that one should make an image of the system partition.

Bye,
Alex

Andy,
it' very kind you asked.
We should all thank you in advance for your efforts.
As it's said earlier is 's very hard to satisfy everybody.

I think lanczos and spline shoulb be fine.
Bicubic , i think it's not hard for you to include it since it seems to be fast enough and bug free.

As i said in an earlier post i think the "tap" thing it's important.
So if you can optimize and debug on of the more "taps" algorithm or implement more "taps" to one of them (lanczos4 or lanczos6) and include the bicubic as is then you 'll have most of us more than happy.

Thank you in advance

Originally posted by vkon
Andy,
it' very kind you asked.
We should all thank you in advance for your efforts.
As it's said earlier is 's very hard to satisfy everybody.

I think lanczos and spline shoulb be fine.
Bicubic , i think it's not hard for you to include it since it seems to be fast enough and bug free.

As i said in an earlier post i think the "tap" thing it's important.
So if you can optimize and debug on of the more "taps" algorithm or implement more "taps" to one of them (lanczos4 or lanczos6) and include the bicubic as is then you 'll have most of us more than happy.

Thank you in advance

hehe its already obsolent, u can use all modes besides sinc. Btw bicubic, lanczos, spline all use the same general resizer routine and just differ by the taps and coefs (aka sharpen) they use. While bicubic is a 6/4 tap lanczos is a 8/6 and spline a 10/10 tap routine. So in general Spline will be the same like lanczos with 10/10 taps, maybe the coefs differ but i can try get those from avisynth code.
I found a other way to implement the code and i will implement all routines wich use 2-10 taps. Sinc with 32/34 tap's is just to crazy to be usefull and since im fully unrolling the loops i dont want copy them 34 times for y/u and v.

PS: btw some1 know where to look for lanczos4 and lanczos6 code/coefs? I found this but its not a open project (http://www.general-cathexis.com/interpolation.html)

I just found the lanczos4 coef its like ....sin(d*PI/4) for 8 taps and a para x from -4.0/+4.0, i assume that lanczos5 is sin(d*PI/5) for 10 taps and parameter x -5.0/+5.0. I can try implement lanczos4 and 5. lanczos6 should be the same with 6 and 12 taps than, but that will make lanczos6 2 times slower than lanczos3.

Andy,

Thanks again for your efforts. One day when we finally decide to establish AVS Forum HPTC hall of fame, you have my vote.

Am I reading your post correctly that resize code optimization is done and ready for use?

Originally posted by vpopovic
Andy,

Thanks again for your efforts. One day when we finally decide to establish AVS Forum HPTC hall of fame, you have my vote.

Am I reading your post correctly that resize code optimization is done and ready for use?

not yet, i have done 4 diff. version's for the y, v, u upscale and finaly decided wich version would be best (fastest) now i have to do a lot of copy/paste work and test them again.

PS: btw most work was done by all the mplayer/xvid/ffmpeg devs and for sure MILAN, i just profile the code and work on hotspots. For example my first own version of the inner resizer routine was crappy slow, than i took the time and tryed to understand what Michelle is doing in his org. mmx code and understand why he was doing it this way. I just adapted this than for my sse2 code and decided what else i can do for speed. So mainly im just manipulating the work of others, but all credits should go to them. (hehe and i like to post more often)

ah it seems i also found all pixelbuffer entrys wich control how far u can resize and u should be able to resize to 4000x4000 than. I also found the reason why the org. code just allow 2000x2000. The overlaymixer only connects with max of 2000x2000, for higher resolution u need vmr7/9.

Thanks Andy. Milan and other FFDShow and Avisynth developers are definitely on my list, together with Blight (ZP). So you should be in good company.

um i just tested lanczos4-6, seems to work but how can i see if the routine is working correct? I mean on my monitor i hardly see a difference from bicubic to lanczos3.... What i have to do to see if my lanczos4-6 values are realy working and improving the image? Some1 have a test avi with a spot to look and see the differences?

PS: btw what is the "parameter 1-10" slider supposed to do anyway in the settings tab of lanczos?

Andy ,
i don't think you can see differences on your monitor.
If you think your code is ok you could publish the release to have some members do the testing.
Many differences need a bigg screen to be visible.

This thing about the parameter 1-10 i think it's been asked sevral times and no one answered it .
I don't know if you can find out from the code.
Some members posted they don't see any change when they adjust the slider.

Heh, we were hoping you'd know what the 1-10 parameter is. Nobody else seems to!

Originally posted by Jeraden
Heh, we were hoping you'd know what the 1-10 parameter is. Nobody else seems to!


mhh i think it was for choosing the tap deep and parameter range for the resizer, but its actually not used. Maybe i can fix this and use it to choose lanczos 2-10 mode.

I know you guys are mostly (all?) using this for DVDs..... but....

What would you recommend for SDTV captures from a Hauppauge PVR 250 card?

I think some folks use Greedy2Frame or TomsMoComp for deinterlacing, but I'm not so clear on the "most best" processing steps to take when you have relatively blocky MPEGs like those that come from a PVR card.

My thinking is that getting the best deinterlacing (is this TomsMoComp for sdtv captures where sports sometimes show up?) and doing levels may be all that really matters given the source material?

TIA,
Stuart

AndyIEG I need some things to be cleared please ;)

I use ffdshow for 2 matters.

- 1st one is DVD with WinDVD 6 filtes via ZP.
In order to be used with ffdshow, WinDVD 6 outputs YUY2.
I guess I should use YUY2 as ffdshow output.

- 2nd one is WM9 HD via ZP.
ffdshow says it gets YV12 as input, so I guess I should use YV12 as output.

Is this right ?
Will there be a difference (and what difference) if I use YUY2 instead of YV12 and vice-versa ?
I've read that ffdshow uses YV12 internally, and that would be the reason why ffdshow is faster with YV12... is it true ?

I guess the general question is... should I use YV12 or YUY2 ? :)

Originally posted by AndyIEG
um i just tested lanczos4-6, seems to work but how can i see if the routine is working correct? I mean on my monitor i hardly see a difference from bicubic to lanczos3.... What i have to do to see if my lanczos4-6 values are realy working and improving the image? Some1 have a test avi with a spot to look and see the differences?
Can't you use the Grab feature of ffdshow? When I use this, it gives me images at the resized resolution, and not the resulting overlay resolution of my monitor (after the graphics card gets hold of it). Hence, a non-resized image is small, while a resized one is big. This helped me to determine that Gradual Denoise is almost as good as Denoise 3d (after applying a Lanczoz resize). Considering my system can do the Gradual Denoise and Resize, I'm going that direction for now.

PS: btw what is the "parameter 1-10" slider supposed to do anyway in the settings tab of lanczos?
I was wondering that myself. Can't you trace the source from slider control through to the application of its setting?

-=> Mike Hanson <=-

To get one thing straight, im NOT a videophile HTPC profi.... i dont even need ffdshow realy for watching videos on my pc. I just have fun to learn and work on something that some1 else use and need.

For the video HTPC Owen and co is your men :)

All i know is that yv12 is faster but on my PC monitor i have to limit the level range to 16-240 with a gamma correction of 1.2-1.3 to have better colors and stuff.

But look some pages before there was a big discussion around this.

Several nights ago, I finally got around to setting up ffdshow and Zoom Player on my HTPC. Tweaked and fiddled around till 5 AM (posted a few pages back - thanks for replies).

Anyway, that night I only tested a few DVDs - Star Wars Ep 1, Ep 2, HP & Ch of Secrets, then just quickly 5th Element. I was not able to get SW Ep 2 running after a few tries, and I think I remember seeing a Macrovision error message or something.

Then last night, I was playing around with different DVDs from the Reference DVD list that I happened to have, plus some others - Shrek, Lion King, Finding Nemo, Dark Crystal, SW Ep1, etc. However, on all of these the menus did not work. It would come up black, with the audio going. Right-clicking to try and get into the Chapter or other menus did not work, only switching the audio. I would have to exit the program and bring it up several times, trying to get into the actual movie to see anything. Then, I wanted to actually watch SW Ep 2, popped it in - and to my surprise, it came up fine, and even played through all the intros and menus fine, let me use the mouse to move around menus, and played the movie great (looks amazing, btw).

I was wondering if anyone could help me with this menu problem, as in, how to make sure menus work correctly at least some of the time. (I remember reading here that sometimes the menus don't work anyway)

Here's my system and setup.
HP Media Center P4 2.8Ghz, 512 RAM, NVidia GeForce 5200 Ultra 128 RAM, out VGA to Samsung HLN4365W DLP set. Zoom Player running ffdshow, with resize to 1280x720, unsharp mask, some NR/Blur settings, some picture settings (gamma/luminance, etc.).


BTW, I was wondering what people meant by "having xxx filter before yyy setting" - it just dawned on me -- is that what the order is on the left side of ffdshow? Moving options with the arrows up and down to set the order they are applied in? If so, I'll have to go back and make sure I have the right order. (--> resize, unsharp mask, blur/nr, gamma/luminance?)

Thanks for any help.

With your system, as similar to my own HP Pavillion P4 2,5Ghz 512RAM and Radeon 9000pro, my settings (with Theatertek player) is;

3D denoise L: 0,5 C:1,0 T:5,0 HQ
Resize Lanczos 1280x864 Luma sharpen: 1,5 Croma: 0,0
Overlay enabled
Codec: YUV12 Overlay mixer enabled

Works fine with all region1 and 2.

Using Powerstrip:
1280x864@50Hz PAL
1280x720@48Hz NTSC

Thanks.

BTW, is "works fine with all region 1 and 2" because of a fix to your TheaterTek software, a modified DVD player, or something else?

Originally posted by AndyIEG
All i know is that yv12 is faster but on my PC monitor i have to limit the level range to 16-240 with a gamma correction of 1.2-1.3 to have better colors and stuff.
But look some pages before there was a big discussion around this.
I read that already but it's not clear enough (for me at least)...
If someone could really explain what benefits one would have to use YV12 or YUY2 as ffdshow output, it would be great :)

Originally posted by cyberbri
Thanks.

BTW, is "works fine with all region 1 and 2" because of a fix to your TheaterTek software, a modified DVD player, or something else?

You must import an registry settings when switching PAL/NTSC.
ntsc_pal (http://rennfast.dns2go.com/Public/HTPC/ffdshow/ntscpal.zip)

PAL/NTSC? The Region 2 DVDs I'm worried about are from Japan, and they work the same as Region 1 (NTSC?), except the region code is different. Will this still work, importing the registry settings when I just want to switch the region code? Will doing the PAL one change my system to PAL?

Originally posted by Owen
It does not hurt to try disabling HT on your system, just to see if it helps, but don’t assume it will be faster.

I wish I knew that it wouldn't help performance before I spent a week setting up and stresstesting my best install to date! :( But as long as disabling hypterthreading won't reduce ffdshow performance it's OK, since that's the only thing that worries me. I've got enough power to do anything else I want to do with that computer. But should I do a new install with hypterthreading enabled? Questions, questions... :confused:

Originally posted by Mastiff
But should I do a new install with hypterthreading enabled? Questions, questions... :confused:
Yes, if you use Windows XP, you shouldn't normally disable hyperthreading. Hyperthreading is the only real edge the P4 has, and Intervideo's filters make good use of it, not to mention the other background tasks always running. For example, on my P4 system, during DVD playback with Intervideo and ffdshow, I get 60% utilization on CPU 1 and 20-30% on CPU 0.

The only reasons I'd disable it are (a) if you are going for extreme overclocking, as HT can hold you back; (b) you have some hardware of software that crashes with HT (very rare); or (c) you don't use Windows XP.

If you change HT settings (from on to off or vice versa), you should reinstall the OS, or you may experience stability or performance problems.

Mastiff,

I spent some time today doing some more hyperthreading tests with Andy's latest build. I used VMR9 with Spline 1440x960 (luma sharpen at .47) and measured total number of frames in 16 minute periods and also the average frame jitter in milliseconds. I used YUY2 out. Then I added Denoise3d with .5 .5 5 HQ and did the tests again. Lastly I used Spline at 1920x1080 with no other filters to overload the machine. I used Raiders of the Lost Ark which is 2.1 to 1 widescreen as my test video. This has a lower processor requirement than 4:3 sources due to not processing the black bands.

To get to the point - it was pretty much a push between hyperthreading and no hyperthreading with a slight advantage to hyperthreading with a heavy, but not overloaded, CPU. When overloading the processor the FPS dropped to 17.92 with hyperthreading and 17.97 without giving a slight edge to non hyperthreaded operation in this way too overloaded non-watchable situation.

My machine was able to keep up with the 24FPS material just fine with Spline only and hyperthreading made no real difference except HT jitter was 2ms and non-HT jitter was 3ms. When I added denoise3d the FPS dropped to 23.92 for non-HT but stayed at 24 FPS for HT! Jitter was as above.

With real video and the CPU not overloaded the performance is slightly better with hyperthreading enabled on my machine. You mileage may vary.

Charlie

Now I can resize!!!

Observations:

1/ With system as it was I turned on Hyperthreading (system was installed with it on, and following comments here it had been later turned off). With HT on I could actually get to 2 x DVD (1440 x 1052) bicubic. Not Lanzcos though.

2/ Pulled out Sapphire 9000 card, put in Gigabyte Radeon 9600 (R96128D) card. This is affordable, and has heatsink not fan. Bam! Resize happy at 2 x DVD even with spline. No dropped frames.

Problem fixed then.

For ref I could see when I was in trouble as although CPU reading may have been looking OK I would notice Reclock couldn't determine the video frame rate. It stayed red even though refresh was a mutliple of frame rate. Dropping resize down fixed that. Now it doesn't matter.

gazzagazza, most people will say that I hate to say that I told you so, but I love to say it! I told you so, I told you so, I told you so, I told you so, I told you so, I told you so, the graphics card had it!

nm88, I'll take door number a! I would say that running a 3.0 @ 3.8 on air cooling is rather extreme. Wouldn't you? I know may who aren't able to go furhter than 3.4-3.5 on the same processor and motherboard, probably because of HT.

Charlie, that one I really appreciate! I would like to think that you did it only for me, but I'm sure you have benefits of it yourself. And other people will to. Since the difference is so minimal, and since my overclock probably wouldn't at all be possible with HT, I will leave it off. :cool:

I don't get that gazzagazza. I can resize to double dvd res w/lanzcos on my system with a 9000 pro. How exactly does the video card affect it? - It's all CPU as far as I understood.

Hi sorry if this as been answered before,this thread is so long

I am running a AMD 64 bit cpu and Iam not sure which version of FFDshow to use, do I need the SSE2 version or just the plain version :confused:

Steve

AMD 64 and Pentium 4 processors both support SSE2. I use it on my AMD 64 to good effect.

Read the last 10 pages or so of this thread to get the current thinking on various configuration issues.

Originally posted by cyberbri
PAL/NTSC? The Region 2 DVDs I'm worried about are from Japan, and they work the same as Region 1 (NTSC?), except the region code is different. Will this still work, importing the registry settings when I just want to switch the region code? Will doing the PAL one change my system to PAL?

The registry settings does only handle NTSC/PAL. If you run an NTSC movie (region1 or 2), you simply just import that registry setting.

The settings for NTSC is then active until you wanna run an PAL movie, THEN you just import the registry setting for that. You still must run the regionkiller for region change in WinXP.

btw. I'm using an ATI Remote wonder remote, and has assigned 2 of the programmable buttons for the registry import...

You can use Anydvd to defeat region protection AND drive powerstrip to change res + merge reg entries for PAL/NTSC automatically - using bat files.

DvdAutorefresh will do it too, so will Girder I think.

Originally posted by Mark_A_W
I don't get that gazzagazza. I can resize to double dvd res w/lanzcos on my system with a 9000 pro. How exactly does the video card affect it? - It's all CPU as far as I understood.

Mark

I was wondering the same thing - is it as simple as whether Overlay or VMR9 is being used by any chance. I was assuming Overlay.

Bill

Maybe some1 know this, im working still on the resize code wich does massive multiply op's. Now i look into the AMD64 ref. and a multiply can be done every 2 cycles with a latency of 4 cycles for the result (pmulhw). In the IA-32 Intel ref. paper the multiply needs 9/8 cycles to be rdy with the same throughput.
I also see that this is on CPUID 0F3n and 0F2n P4 machines, right next to them is the 0x69n refered as Pentium M and the latency of the same instruction is just 3+1.
Now i wonder is the Pentium M faster or do both CPU's have diff. cycles per clock?

Andy,

Don't know about Pentium M, but just want to clarify P4 vs. AMD64 situation with SSE2 instructions and multiply (which is probably where the most of the CPU intensity for resize is, right?).

You are saying that AMD64 can perform multiply operation every second cycle with latency of of 4 cycles. So let's see how much it takes for AMD64 to render 5 frames. My math will probaby be wrong, so please correct me:

1 - 1
2 - 1+4+1=6 (i.e. cycle for 1st frame, latency 4 and new cycle for 2nd frame)
3 - 6+4+1=11 (i.e. cycles for 1,2nd frame, latency 4, cycle for 3rd frame)
4 - 11+4+1=16
5 - 16+4+1=21

Now P4

That should be simple, latency of 8 cycles between two multiply operations (right?).

1 - 1
2 - 1+8+1=10
3 - 10+8+1=19
4 - 19+8+1=28
5 - 28+8+1=37

So if AMD64 uses 21 cycles for 5 frames, while P4 uses 37 cycles for 5 frames, that gives us ratio of 1.76 i.e. P4 needs 76% more cycles to perform calcs for the same number of frames. If you run calc for more frames it looks like the ratio increases until 1.8 and then stays there. At the same time, if you compare AMD64 and P4 ratings, P4s have about 41% clock speed for the same "rating" (i.e. 2.2 Ghz AMD "3200+"). So, if my calc is right (probably unlikely) it looks like with SSE2 instructions AMD64 could do significantly better than P4. For example, based on the (questionable) calc above, it would take 4.7 Ghz P4 to do the same multiply work as 2.6 Ghz AMD64 (factor 1.8).

That is theoreticly right, but we have to know some more facts. All instructions or also called macro ops will be break down into micro ops in the decoding and reordering phase. Now we also would have to know in how many micro ops each cpu breaks the multiply. The other problem is each cpu has diff. cycles per clock. I think the AMD64 has 8-9 per clock, i heared that the P4 has much more. The question now is how many micro op's per clock we can do.

But i see since we arnt CPU designers all i can do is to shedule the multiplys as best as i can and keep the dependency chains short.

But thx for the math :)

PS: ah the 8 cycles so called latency means that if u start a multiply the result will be finished 8 cycles later, so u have to do something else beetwen the multiplys if u want to do something with the result.
The main problem is to keep the cpu busy while u wait on some results, thats why shorter latency also means faster execution in long dep. chains in wich each value waits for the previous result.

Example:

1: a*b store in a
2: ----
3: b*c store in b
4: ----
5: d*e store in d
6: ----
7: f*g store in f
8:
9:a + 1

on the P4 i can do "a + 1" in the 9 cycle, while on a AMD64 i can do it at cycle 5

U can sure start a multiply every 2 cycles if the 2 operands not depends on a previous value wich is not rdy.
So

1: a*b store in a
2: ----
3: c*a store in c
4: ----

this will result in a big stall since c*a cant be done yet.

Originally posted by Mark_A_W
You can use Anydvd to defeat region protection AND drive powerstrip to change res + merge reg entries for PAL/NTSC automatically - using bat files.

DvdAutorefresh will do it too, so will Girder I think.

Could yoy please guide us in how that could be configured?

My setup:
PAL.REG (registry import for reg.2)
NTSC.REG (registry import for reg.1)

Powerstrip:
1280x864@50Hz (PAL)
1280x720@48Hz (NTSC)
Created custom app.lunch for PAL/NTSC where they points to the 2 registry files. However Powerstrip does not switch rez when i klick on the app. settings...?

Assigned registry import for PAL/NTSC on custom buttons on the ATI remote ok.

This is what I want it to do:

When pressing the "A" button witch is assigned for NTSC on my remote, I would have the registry file imported, AND powerstrip to switch to 1280x720@48Hz.

Cannot get this working..:confused:

Silly question, but what is the difference between ffdshow, and ffdshow SSE2?

I just saw that there are newer versions than I'm running, on the http://mitglied.lycos.de/ieggei2/ffdshow/ link that was posted.

Originally posted by cyberbri
Silly question, but what is the difference between ffdshow, and ffdshow SSE2?

The SSE2 version is optimized for SSE2 compatible CPU's (AMD64 / P4).

ahh, thank you. Guess I need the SSE2 version.

Rolf, PowerStrip supports command line, so you could create a regular DOS (anybody remember DOS?) batch file that first imports the registry and then c:\programfiler\powerstrip\pstrip.exe /P:PAL (you need to rename your PAL setting from 11... to simply PAL or use the settings name, but remember to stay within legal DOS characters). I do that with Girder.

Guys, that "nothing has changed" scare has gone over. I did a couple of hours of tinkering today (while listening to the swedes' incredible match with a goal five minutes from finish) and even though I SUCK big time in especially focus I could easily see that both 1152x864@75 and 1536x864@50 (which seemed better than the same setting @ 75) made the differences from the extreme resize pop out. I really didn't see that much change from denoise 3D but I suspect that will only show itself when the projector is correctly adjusted. So now I need somebody to redo my setup - I see no point in spending hours and hours in the HT never even getting close to what the BG can do. And when that's done I will compare the settings and find out which one I will use.

Oh yes, I saw tearing! Too bad, though, because VMR9 seemed more "lifelike" than overlay, but the difference was slight. :cool:

Rolf, I will make a new thread.

Originally posted by Mark_A_W
I don't get that gazzagazza. I can resize to double dvd res w/lanzcos on my system with a 9000 pro. How exactly does the video card affect it? - It's all CPU as far as I understood.

Well, someone may be able to explain it. I don't know the reason. Read back a couple of pages and you'll see others telling me the bottleneck must be the video card. I'm using Overlay. The difference between the results with the 9000 and the new 9600 are substantial. I no longer seem to need reclock to get a smooth replay. This suggests everything is much happier?

Originally posted by Mark_A_W
I don't get that gazzagazza. I can resize to double dvd res w/lanzcos on my system with a 9000 pro. How exactly does the video card affect it? - It's all CPU as far as I understood.

Consider what must happen when you resize much larger than the final image, then ask the video card to downsize it to the final resolution:

The CPU must compute and write the image data. This can be done in one of two ways:
a. The CPU writes the data into main memory and the card transfers it to its memory before processing, or
b. the CPU writes the large scaled output directly to video memory.

Either way, the path to video memory is critical to getting a large image to the point where it can be scaled back down to actual display size.

Modern cards have both faster and wider memory which permit either strategy to work with larger images than older cards.

Originally posted by Charles Black
Owen,

That will do a good job - much faster than we used to use. I don't think that a BW of less than 10Mhz will be very good for video. I have an old Tek 323 that I think is real slow that I'l put on the video tomorrow to see what it looks like. If your scope has delaying sweep and calibrated vertical amp you can do more than just ringing. All you need is a good probe. If it is really simple and only has a single sweep you can sometimes use tricks to delay your trigger a bit too. What kind is it?

Charlie

Sorry about the slow replay Charles.

My scope is a full featured analogue dual trace 60Mhz Tektronics unit with high quality compensated probes, dating from the late 1980’s. It’s a left over from my HiFi audio electronics design and testing days.
I can’t remember the model number but it does have delaying sweep and a calibrated vertical amp.
I think the last time I used it was about 2 years ago, to repair a power amp for a friend.
This scope was highly over spec for audio work, but I think that it may be a bit marginal for video. Especially if I want to look at square wave response for ringing or overshoot.
I have never tried to look at video waveforms, so finding out how to sync the scope will be a learning experience.


Regards,

Owen

Owen,

Your scope is plenty good enough enough for video use. I was just using the bandwidth limiter on my scope and at 50mhz traces were about the same as full BW. At 20Mhz test pulses were the same and overshoot was still visible even though noise was much reduced. At 50MHZ one of the test patterns where alternating black and white bars get narrower and narrower the narrowest were at about 80% amplitude and the rest were full amplitude. I am too lazy to crank up my fast scope to compare - I only use it for rise time measurements and glitch detection at this point.

I hope you have BNC terminations on you video cable but if not it is not too expensive to get an adapter or use a double 15 pin to BNC setup. Don't try to use the scope probe in the 15 pin connector (probably cast anyway) as there is not enough room.

Charlie

hi there all.
this is my first post regarding ffdshow so be please be gentle.
after a couple of weeks reading of this thread and trying to take it all in i have decided to give this program a go.
i was wondering if anyone with a similar set-up as myself can tell me if they've managed better settings than i have managed to get without any stuttering in the film?
my htpc specs are as follows:
amd athlon 2600+
theatertek
radeon saphire 9600 (non pro)
ffdshow (latest version 20040607a i think)

the best settings i can achieve are:gradual denoise set at 30
and then
dscaler sharpness filter set at 128 with fast memory access enabled.
are these good settings or is there much better and more advisable ones recommended?
i've tried resize and lanczos options but my htpc doesn't seem to like them at all.
any info would be greatly appreciated.
paul

You should try the following with that hardware setup:

In this order:

- Codec RAW only (all supported)
- denoise 3D with Luma: 0,50 Croma: 1,00 Time: 5,00 Enabled HQ.
- Resize Lanczos (1280x720 for my part) Luma sharpen 1,4 Croma sharpen 0,0
- Output UV12 only.

denoise 3D is allot better than gradual denoise.

Use the optimized ffdshow available here. (http://mitglied.lycos.de/ieggei2/ffdshow/)

I just started using ffdshow, and I've noticed that it causes significant color banding, especially in red areas. It occurs regardless of what settings I use in ffdshow; in fact, I can turn everything off and I'll still get the banding. Also, it occurs both in overlay mode and VMR9. So basically, if I run Zoomplayer without ffdshow, it's fine (albeit without the resizing and Denoise3D that I've grown to love), but when I add ffdshow as a post-processing filter, regardless of any settings, I get banding. Am I doing something wrong?

Flanger

P.S. I just remembered: I tried every single colorspace output mode, but each one still results in banding (although to different degrees). I also tried checking and unchecking 'use overlay mixer' under the output tab, but that doesn't seem to do anything.

Originally posted by paul1672
the best settings i can achieve are:gradual denoise set at 30
and then
dscaler sharpness filter set at 128 with fast memory access enabled.
are these good settings or is there much better and more advisable ones recommended?
i've tried resize and lanczos options but my htpc doesn't seem to like them at all.
any info would be greatly appreciated.
paul

Before we go too much further, please describe "stuttering" What exactly are you seeing?

Also start out with just resize, no other options, and use bicubic, and work up to a resize level that causes a problem. Post that... that'll give an idea of how limiting your PC is...

You could turn on the OSD as well and see what it says about your ffdshow performance.

Originally posted by vpopovic
gazzagazza,

I'll think about your issue, sure strikes as odd. I never ever had 9000 in my system, so I was wondering if that could be the problem. What happens when you resize to 1440x960? When your filters work well there is really no need for Reclock.


Lets digress here and talk briefly about reclock. With my new graphics card and much improved resize ability in ffdshow I'm still getting a stutter now and then. Best example is actually outside Zoomplayer in Windows Media Player running a WMVHD clip (Flight). Its just replay of a 24fps clip on a system with a 50Hz refresh. On the opening sequence as you fly down the river the forest on the righthand side shows a small step at regular intervals, maybe once a second... Add reclock with framerate set at 24fps and the motion is completely smooth.

Any comments anyone?

Originally posted by gazzagazza
Lets digress here and talk briefly about reclock. With my new graphics card and much improved resize ability in ffdshow I'm still getting a stutter now and then. Best example is actually outside Zoomplayer in Windows Media Player running a WMVHD clip (Flight). Its just replay of a 24fps clip on a system with a 50Hz refresh. On the opening sequence as you fly down the river the forest on the righthand side shows a small step at regular intervals, maybe once a second... Add reclock with framerate set at 24fps and the motion is completely smooth.

Any comments anyone?

Reclock will "speed up" the framerate of 24fps material to match a 50Hz PAL refresh. Since it's only 4% faster most people won't notice that the audio will be slightly off pitch.

I have also been trying reclock lately (PAL sources) as i was getting almost random stutters where over say a long pan it would stutter once, but other times on a diagnal pan it may stutter a few times.

Using Reclock on "auto(best)" with my refesh rate set to 50hz (as always) has not helped. Infact reclock makes it worse with constent tiny sequential stutters.

While my refresh is 50hz, it never seems to be "exactly" 50hz. Ie recock may read it as 50.1 for example. I did try fine tunning it in powerstrip ang got it to 49.998hz but didn't seem to help (and seemed lost on next reboot).

Have very little experience with relock, not sure what else can be tried.

Spoonfed, is your display device actually displaying 50hz?

yeah, 50hz. (infocus X1)

i just figured perhaps one could never get exactly say 50hz due to inperfections on the system, clocks, video etc etc hence 4.998 or 50.100 etc
(9600pro by the way)

Just to put my mind at rest, could you try it on a CRT monitor?

Then you'll know whether the projector is partially the problem, or it's just the PC.

when i say "stuttering" i mean the film is pausing very briefly every so many frames.
i will give it another go with your suggested method and post results later.
many many thanks
paul

right,i have tried rolf hults suggestion and it is running ok,only thing i'd like to know is did rolf mean output to be YV12 only as i couldn't see the option for UV12.
i also turned on the osd and it was flashing between 67% and 100% so i guess that means i'm very near my limit.
i assume i have the settings in the correct order?
denoise 3d first and then resize?
one other thing i'd like to ask,i am displaying the film using a sony hs-10 pj which displays at 1366x768 resolution.
does this make any difference as i'm only resizing to 1280x720?
does this mean i am no longer 1:1 pixel matching?
sorry to sound thick but i'm new to all this.
many thanks for the help.
paul

Paul, are you PAL or NTSC? Does it work OK and smoothly without ffdshow?

Personally I don't see any benefit from anything other than resize... just set it for 2 x DVD res, and forget all the other stuff. YV12 output.

Once you start worrying about those little stutters you lose the point of it all as the enjoyment of what you're supposed to be watching is lost.

i'm im the uk but play mostly ntsc films,i do have a few pal ones as well though.
it plays flawlessly without ffdshow.
only weird thing is,when ffdshow is installed i can't play any dvd from the drive without experiencing break up in playback.it has to be put on the h/drive first to get smooth playback but that's not a major problem.
seems to be a problem with theatertek as if i then uninstall ffdshow the problem still exists.
oh well.
all seems to be running well with the settings rolf hult mentioned.
i have tried just resizing to x2 dvd size (1440x960) but i get the usual problems.
so,i guess the settings rolf has mentioned must be my optimum.
many thanks for the advice.
paul

Originally posted by gazzagazza
Once you start worrying about those little stutters you lose the point of it all as the enjoyment of what you're supposed to be watching is lost.
I'm not really sure what you're suggesting here. Are you saying we should just ignore the stuttering? That's equivalent to saying, "Listen to this fantastic recording of Beethoven's 9th. Oh, and ignore the rhythmic hiccups that occur. (They should only happen once every minute or two, although they might come as often as once a second.) Instead, just immerse yourself in the glory of the music."

Rhythmic glitches in playback completely ruin my enjoyment! I suppose that's why I demand proper "PRaT" (Pace, Rhythm and Timing) from my audio gear, much more than proper tonality. If the flow isn't correct, then it doesn't matter if it "sounds like the real thing".

If I cannot get ffdshow to run virtually error free (or at least studder free), then I'll just have to turn it off. I'm willing to live with a fuzzy image before a studdering one.

If your statement above meant something else, then please ignore my rant. :)

-=> Mike Hanson <=-

Quote paul1672

“right,i have tried rolf hults suggestion and it is running ok,only thing i'd like to know is did rolf mean output to be YV12 only as i couldn't see the option for UV12."

Yes he did. Unselect everything except YV12.

Quote paul1672

“ i assume i have the settings in the correct order?
denoise 3d first and then resize?”

Yes

Quote paul1672

“one other thing i'd like to ask,i am displaying the film using a sony hs-10 pj which displays at 1366x768 resolution.
does this make any difference as i'm only resizing to 1280x720?”

Yes.
Vertical resize is more important then horizontal resize, especially with NTSC DVD’s that only have 480 lines to start with.
There is no rule that says you have to resize horiziontal and vertical by the same ratio. You video card will resize everything back to your desktop resolution anyway.
I would try something like 960x960 for NTSC. That will be the same system load as 1280x720 but may look better. Give it a try.

Quote paul1672

”does this mean i am no longer 1:1 pixel matching?”

No.
As long as your desktop resolution is set to your displays native resolution you are still 1:1.



Regards,

Owen

___________________________
The FFDShow resize-sharpen dude.

ok,tried the 960x960 setting and that too looks good.
what would be a good setting for pal films?
i tried doubling the pal vertical size (1144) but its a no go.
and if i leave it at 960x960 that also wont play without problems.
anyway,i finally feel like i'm getting somewhere.
many thanks
paul

Someone asked me for DMO_V InterVideo Abstract registry tweak and here it is. Just extract and run abstract.exe inside the attached file.

Hi:

Just wondering if there was a way to make ffdshow support use on secondary monitors. I use Theatertek with ffdshow. I have to set it up so that my tv is my primary monitor or else it just hangs when it loads.

Now, setting it up this way wouldn't be a problem - save for the fact that programs open on my tv by default, as it is now the primary monitor. This is an issue because i have to switch inputs all the time if I want to look at something like task manager.

Andrew at Theatertek informed me that this was an issue with ffdshow and not with TT so I'm hoping that someone who uses both - or is just very knowledgeable - would know what to do to fix the issue. I love the benefits of ffdshow's post processing, but will have to turn it off just for the sake of ease of use with the computer.

Thanks,
Jeff

Originally posted by boxsoft
I'm not really sure what you're suggesting here. Are you saying we should just ignore the stuttering? That's equivalent to saying, "Listen to this fantastic recording of Beethoven's 9th. Oh, and ignore the rhythmic hiccups that occur. (They should only happen once every minute or two, although they might come as often as once a second.) Instead, just immerse yourself in the glory of the music."

-=> Mike Hanson <=-

No, I'm suggesting that if you tweak for performance, but then introduce small stutters etc in your quest, you will ruin your enjoyment. Less may be more...

Regarding the problem playing DVDs from the DVD drive, have you checked to see you are running in DMA mode on the DVD drive?

isn't DMA mode enabled default on xp?

Originally posted by bbq@KL
Someone asked me for DMO_V InterVideo Abstract registry tweak and here it is. Just extract and run abstract.exe inside the attached file.

I'm wary about running unknown exe files on my PC. What exactly does this do?

isn't DMA mode enabled default on xp?

Not usually. Especially if your chipset can use the Intel Application Accelerator software (a lot of them do). I believe that software controls whether it is enabled/disabled automatically. If you don't have it installed and your chipset can use it, you are most-likely running A LOT slower than you should be. Was more noticeable on the earlier P4 boards though.

Originally posted by Arcanum
Hi:

Just wondering if there was a way to make ffdshow support use on secondary monitors. I use Theatertek with ffdshow. I have to set it up so that my tv is my primary monitor or else it just hangs when it loads.

Now, setting it up this way wouldn't be a problem - save for the fact that programs open on my tv by default, as it is now the primary monitor. This is an issue because i have to switch inputs all the time if I want to look at something like task manager.

Andrew at Theatertek informed me that this was an issue with ffdshow and not with TT so I'm hoping that someone who uses both - or is just very knowledgeable - would know what to do to fix the issue. I love the benefits of ffdshow's post processing, but will have to turn it off just for the sake of ease of use with the computer.

Thanks,
Jeff



I will preface my following comments by saying that I have never used TT, but I am familiar with the Sonic decoder the TT uses, although it’s a long time since I used it.

Users of Zoom Player can use FFDShow on multiple monitors without problems.
So to say that FFDShow is the problem and not TT is misleading IMHO.

There are a couple of issues that may limit TT’s multiple monitor support.
The first is the video overlay that TT uses. Normally the overlay is limited to the primary monitor only, and only one application can access the overlay at a time.

The second reason is that FFDShow forces the DVD decoder into software decoding mode in stead of DXVA hardware accelerated mode.
If DXVA mode is used, the limitation of the overlay to only allow display on the primary monitor is avoided. But with FFDShow in use, you are stuck with the limitations of the overlay system.

Unlike TT, Zoom Player can use VMR9 to render the video.
VMR9 can display on multiple monitors simultaneously, be they primary or secondary and is not limited to a single application so multiple applications can be displaying video on multiple monitors simultaneously.
VMR9 does not need DXVA for this to work so FFDShow, that forces the DVD decoder into software mode can still work on multiple monitors.

That is the situation as I understand it.


Regards,

Owen

___________________________
The FFDShow resize-sharpen dude.

i got round the dual monitor problem by using a program called ultramon.
it allows me to play a movie on my primary screen (a projector) at 1366x768 and display the windows desktop and open files on my secondary screen (an lcd tv) at 1280x768 both at the same time.
take a look here:
http://www.realtimesoft.com/ultramon/
hope this helps
paul

i checked to see if dma is enabled and it definately is on both ide channels.it says ulra dma enabled on both of them.
guess it must be something else stopping the dvd drive from playing the movie.just weird that it was playing fine from the drive when i had done a fresh install of everything right up until i installed ffdshow.
like i say though,not too bothered.
cheers anyway.
paul

Hi:

Owen - thanks for that answer, it made a lot of sense and answered my question almost exactly. I was blaming ffdshow because Andrew - on the TT forum told me that it was a limitation of ffdshow. Your answer gels more with what is happening. I've gotten around it by making my tv the primary monitor only when I want to use TT. this is one extra step - and kind of a pain in the a$$ but is worth it for the ffdshow processing. I couldn't go back to normal display now if I tried. I would love to try Zoomplayer but have only just bought TT and can't justify spending more money.

While I am on the topic - is there a way to use ffdshow with the wintv2000 application that comes with the Hauppauge pvr250 tuner card? I ask because neither Beyond tv or Sage TV work as well as the wintv for picture quality on my system - they are both inferior and don't work reliably either for me.

Thanks,
Jeff

Originally posted by paul1672
i got round the dual monitor problem by using a program called ultramon.
it allows me to play a movie on my primary screen (a projector) at 1366x768 and display the windows desktop and open files on my secondary screen (an lcd tv) at 1280x768 both at the same time.
take a look here:
http://www.realtimesoft.com/ultramon/
hope this helps
paul

For people with Nvidia cards, Ultramon functionality is built into the drivers so Ultramon is not required.
You can swap primary and secondary monitors with a keyboard shortcut or when a specific application is launched.

Regards,

Owen

___________________________
The FFDShow resize-sharpen dude.

Arcanum: With TV capture cards, use DScaler (www.dscaler.org), it's probably less CPU intensive, and is designed to work well with real-time video input.

speaking of Dscaler, can someone tell me what exactly it does? Im having issues noticing a difference in picture quality when using it w/ "right side only", set at aprox. 100 (though that number's arbitrary for me as I dont know what it represents).

thanks.

short info seems i was right the "parameter" setting in lanczos and bicubic mode was org. to choose the lanczos mode... means u had could choosed already from lanczos3 - lanczos10 (6-20 tap) but like i sayed the parameter was wrong handed and always 0, wich force the standard mode lanczos3 to be used.

I mailed milan and he found the little bugger :) So the funny thing is that ffdshow already always had lanczos3-10 included and also choosing advanced bicubic modes.
For the sse2 version... i just wished i had SSE3...i realy could need the horizontal add/sub instruction included in the prescott... doing it with sse2 is a pain...
So far im just at the point on wich the sse2 code runs as fast as the mmx2 version...so i have to relook all over again. Good news is i found the horizontal lining bug. I also found where the main buffers are declared and u will be able to resize to 4000x4000 (should be high enuff or?) in the next versions.

Wow, nice finds. Sadly my poor little P4 2.8c seems to have maxxed out at a FSB of 235 stable. I was really trying to hit 250fsb so I could apply a couple more clicks of resize or something.

Originally posted by gazzagazza
Regarding the problem playing DVDs from the DVD drive, have you checked to see you are running in DMA mode on the DVD drive?
All my testing so far has been with DVDs ripped to the HDD. When it comes to using the DVD drive, I'm using DVDIdle Pro, which buffers it for me, so I'm hoping that doesn't become a problem.

-=> Mike Hanson <=-

I have a 35 Watt mobile athlon xp 2400 barton stepping IQYHA that I have overclocked 215*11.5=2472.5 that I can run Windvd 6 decoder, 5-tap lowpass deinterlace, denoise 3d, and Lanczos 1920*1440 (4:3 source) with the OSD switching quickly between 67 and 100 utilization. This is with the MMX2 optimized code by AndyIEG, to whom I must thank. I am not done overclocking this, as my idle temperatures are only 37c and load has not been measured, as I have only been running for 2 days. The best part is that this is an $89 chip that is doing what your expensive P4 and A64 systems are sometimes struggling to do.

Originally posted by madpoet
Wow, nice finds. Sadly my poor little P4 2.8c seems to have maxxed out at a FSB of 235 stable. I was really trying to hit 250fsb so I could apply a couple more clicks of resize or something.

That's very strange. One of those should be able to do 250 without any problems. Is this with or without HT? I have my high overclock probably (I haven't tested, but I'm pretty sure) because I'm not using HT.

With HT. I guess I could disable it and see if it helps. Can you just turn it on and off without hosing Windows XP?

I'm at 1.6 vcore and 235FSB with GAT set to AAADD. vdimm set to 2.8, I might try 2.9. If I bump to 237 I start failing Prime95. I'm going to try downrevving my BIOS 1 version (I was told 1.5 might be better than 1.6) and a couple other tricks I've read about.

Owen (and anyone else up for commenting):

Your one of the ffdshow gurus and I have a couple questions. I have done my due diligence but find a lot of contradictory statements (maybe all being true depending on circumstance).

The popular feature usage seems to be:

Resize (Bicubic or Lanczos)
Denoise (gradual or 3d)
Sharpen (unsharp mask)
YV12 output

Since horsepower is a problem for most of us, what would be the order of priority of features / filter? What I mean is the features (filter) giving the most bang for the buck (in this case Buck ($) = CPU cycles).

I have seen some put denoise and sharpen before resize and others after. It seems horsepower would be conserved to put resize last but perhaps the denoise and sharpen effects are muted then.

You have stated that vertical resize is more important than horizontal. My DLP RPTV is native 1280 x 720p. Would it make some sense to resize to say 1280 x 1440 to get good resize result while conserving some horses?

I have seen two popular strings for 3d denoise: .5, .5, 5, HQ and .5, 0, 5, HQ. What is your preferrence?

For Lanczos I have seen [Luma sharpen = 1.4, Chroma sharpen = 1.4] and [Luma sharpen 1.4, Chroma sharpen = 0.0] as two popular configs.

I have the sense that the range of optimal ffdshow settings and even which features to use may have a big divide between digital (i.e., DLP / LCD using DVI) vs. analog (i.e. CRT using VGA) displays. Something that works well for one class may not do much or even be detrimental for the other.

DFA

Originally posted by AndyIEG
short info seems i was right the "parameter" setting in lanczos and bicubic mode was org. to choose the lanczos mode... means u had could choosed already from lanczos3 - lanczos10 (6-20 tap) but like i sayed the parameter was wrong handed and always 0, wich force the standard mode lanczos3 to be used.

I mailed milan and he found the little bugger :) So the funny thing is that ffdshow already always had lanczos3-10 included and also choosing advanced bicubic modes.

So do you have any plans to fix that bug? If so, in both versions?

Thanks!

milan already fixed it and the parameter will be correct handed to the resizer, not sure if i can fix the horizontal lineing bug for the normal version but it will for the sse2 version. Still tweaking on some problems... its damm hard to do some work if every cicle count since the main loops run millions of times in some seconds and 3-4 more instructions in the asm code already mean 5-15% more cpu usage...

Andy,

Did you say SSE3? Now that's real interesting for us Prescott users. It could be quite a bit faster but we wan't know unless you code it for us. How is your spare time anyway? ;)

Charlie

Originally posted by DFA
Owen (and anyone else up for commenting):

Your one of the ffdshow gurus and I have a couple questions. I have done my due diligence but find a lot of contradictory statements (maybe all being true depending on circumstance).

The popular feature usage seems to be:

Resize (Bicubic or Lanczos)
Denoise (gradual or 3d)
Sharpen (unsharp mask)
YV12 output

DFA

You know what you're going to be told don't you... you have to try it and see what works best for you. Personally I only use resize... at the moment bicubic with a slight luma sharpen... Denoise 3D does nothing for me on my pj... if I set it high enough to notice it is doing anything at all it removes detail (particularly on skin texture). I want the least digital look to what I watch.

gazzagazza:

No doubt you are right. But certainly the order can be nailed down. Interestingly, I was most pleased when using resize only as well but thought I should explore a little to see if other feature were just not correctly implemented by myself.

Can NVDVD 3.0 codec be forced to output YV12 instead of YUV2 for input to ffdshow? Seems I recall Vlad talking about this but no recollection on "how".

DFA

DFA,

Forceware multimedia will output YV12 if you choose "Nvidia Post Processing" in "General>Post Processing" tab.

DFA,
To answer your question.
The best filter setup is always up to the user but as a guide I would suggest the following in order.

Denise3d L0.5, C1.0, T5.0 HQ
Resize (Bicubic or Lanczos) I prefer Bicubic.
Sharpen (Luma sharpen in resize or possibly Descaler sharpen) Do NOT sharpen Chroma as it only increases noise.
YV12 Output

If you don’t have the power to run both Denoise3d and Resize you will have to make your own decision as to which is more important to you.

Regards,

Owen

___________________________
The FFDShow resize-sharpen dude.

Originally posted by gazzagazza
You know what you're going to be told don't you... you have to try it and see what works best for you. Personally I only use resize... at the moment bicubic with a slight luma sharpen... Denoise 3D does nothing for me on my pj... if I set it high enough to notice it is doing anything at all it removes detail (particularly on skin texture). I want the least digital look to what I watch.


Your experience with Denoise3d very unusual.
I find Denoise3d set at L0.5, C1.0, T5.0 HQ dose not affect detail at all, but has a dramatic affect on noise.
You say that you get detail lose but can’t see the dramatic improvement in noise.
That does not make sense.

Regards,

Owen

___________________________
The FFDShow resize-sharpen dude.

Originally posted by Owen
Your experience with Denoise3d very unusual.
I find Denoise3d set at L0.5, C1.0, T5.0 HQ dose not affect detail at all, but has a dramatic affect on noise.
You say that you get detail lose but can’t see the dramatic improvement in noise.
That does not make sense.

Hi Owen, yes its true, I have tried your denoise 3D settings and it makes no difference to a DVD picture quality here... crank up the Luma and Chroma settings until I can see a difference and there is detail lost... all it does is hit the CPU usage. I'm looking at a 100" approx image from a Panasonic L300 pj... and it can produce a pretty good image. DVD source is the limiting factor as can be seen when playing back some of the WMV-HD clips against which DVD pales...

When you say it has a dramatic effect on noise, what sort of source, and what sort of noise?

Noise looks like pixel-size moving dots, and can be found on any DVD. Denoise3d, even at very low settings, takes control of the picture and removes most of these moving dots.

gazzagazza,

I am referring to the “crawling walls” and pixilation on flat surfaces and in shadow areas that is common with DVD and all Mpeg source.
This kind of noise is OBVIOUS to me on ANY display, even with very well mastered DVD’s

I am using the Elecard decoder combined with VERY high resize, 1920x1728 and VMR9.
This combination produces the LEAST noise and pixilation of anything available, vet I can clearly see the affect of Denoise3d and I would not wont to go without it.

Are you saying that you can see detail loss at the settings I suggested?

Regards,

Owen

___________________________
The FFDShow resize-sharpen dude.

Originally posted by AndyIEG
milan already fixed it and the parameter will be correct handed to the resizer, not sure if i can fix the horizontal lineing bug for the normal version but it will for the sse2 version. Still tweaking on some problems... its damm hard to do some work if every cicle count since the main loops run millions of times in some seconds and 3-4 more instructions in the asm code already mean 5-15% more cpu usage...

Hi Andy,

I did not understand if bicubic/laczos parameter bug will be fixed in both normal and SSE2 version?:confused:

Just another fast question: will SSE2 version run on non-SSE2 processor just taking no advantage of SSE" optimization (will it take advantage of your mmx2 opt?) or won't it run at all?

Thanks!;)

Originally posted by Owen
gazzagazza,

I am referring to the “crawling walls” and pixilation on flat surfaces and in shadow areas that is common with DVD and all Mpeg source.
This kind of noise is OBVIOUS to me on ANY display, even with very well mastered DVD’s

I am using the Elecard decoder combined with VERY high resize, 1920x1728 and VMR9.
This combination produces the LEAST noise and pixilation of anything available, vet I can clearly see the affect of Denoise3d and I would not wont to go without it.

Are you saying that you can see detail loss at the settings I suggested?

Regards,

Owen

___________________________
The FFDShow resize-sharpen dude.

No loss of detail at your settings, but no effect on noise either... or perhaps I just haven't been watching noisy DVDs. Give me an example and I'll see if I can get hold of it and compare. One example in my case... we've been watching our way through the Sopranos series on DVD. In series 2 there was a scene in a hospital at night time, and the wall below the nightlight in a room had that small pixel noise on it... (mosquito noise right?) I didn't think denoise 3D made much difference at your sort of settings, gradual denoise did though, but I didn't like it. I decided it was like grain in a silver halide film source... part of the experience. Live with it then. On the other hand denoise 3D was on with slightly higher settings, about 1, 1, 5, HQ when I watched the first of the LOTR trilogy. Frodo's skin looked very digitally smoothed. Turn off denoise 3D and that processed look disappeared. Remember that my screen size is bigger, so perhaps my experience is different to you? Also I don't want to tweak my life away, I want a solution that looks good, and then to forget it and enjoy the movies.

Humm, I cannot see how screen size would have anything to do with this.
I can clearly see the affect of Denise3d on my laptop 15” TFT display.
All DVD’s have mpeg compression related mosquito noise to some extent but it is usually only visible on some scenes.
Look at smooth background objects like walls and doors. Some times it’s so bad it looks like the walls are alive.
Denoise3d set to my recommended settings (L0.5, C1.0, T5.0 HQ) almost completely removes this very aggravating problem.
As long as the Luma setting is kept at or below about 0.5, no detail is lost.

Denoise3d is set and forget. It does no harm and so can just be left on all the time, for all DVD’s and there is no need to play with the settings.

For over compressed DIVX or Xvid video, stronger settings can be used to very effectively remove compression artifacts. It works better then the post processing filters IMHO.

I suppose that if you have never been bothered by the noise, you should consider yourself lucky. Maybe you would be better off not looking for it. :D

Regards,

Owen

___________________________
The FFDShow resize-sharpen dude.

Often darker scenes can show it up more.

As for varying display types. There is some truth there, but perhaps more in "type" rather than "size". As size is relative to viewing distance anyways.

I fun "noise" on the projector much more natural than on the CRT monitor

Originally posted by Michele Spinolo
Hi Andy,

I did not understand if bicubic/laczos parameter bug will be fixed in both normal and SSE2 version?:confused:

Just another fast question: will SSE2 version run on non-SSE2 processor just taking no advantage of SSE" optimization (will it take advantage of your mmx2 opt?) or won't it run at all?

Thanks!;)

bicubic/laczos parameter bug will be fixed in both versions

For now the SSE2 version just run on SSE2 CPUs, the normal still use cpu detection feature and use mmx and mmx2 if its detected.
I was thinking about merging the SSE2 code into the normal version but since im working torward a 64bit AMD64 version based on the SSE2 version it would cause more trouble and debug problems.
I dont think its this hard to choose what version u download... we all do this for drivers and other programms. (the code size is also reduced since the SSE2 version dont need the mmx or mmx2 versions)
The other advantage is that i can compile the normal c++/c code for better sheduling on P4 cpus since the normal version is still a i586 compile.

Originally posted by cetoole
I have a 35 Watt mobile athlon xp 2400 barton stepping IQYHA that I have overclocked 215*11.5=2472.5 that I can run Windvd 6 decoder, 5-tap lowpass deinterlace, denoise 3d, and Lanczos 1920*1440 (4:3 source) with the OSD switching quickly between 67 and 100 utilization. This is with the MMX2 optimized code by AndyIEG, to whom I must thank. I am not done overclocking this, as my idle temperatures are only 37c and load has not been measured, as I have only been running for 2 days. The best part is that this is an $89 chip that is doing what your expensive P4 and A64 systems are sometimes struggling to do.

Try that with VMR9.
I thing that will bring you back down to earth. :D


Owen

Hehe, so very true. And toss in the Elecard decoder as well.

I do get tearing with VMR9. It is very close, and decoding framerate is around 22, but it doesn't quite work, yet. I can use Elecard, but only with vob files. How do you get it to work with encrypted DVDs? I have Anydvd, but it won't connect to the graph.

One thing I have noticed about the Elecard decoders is that it allows me to watch MPEG-1 files without messing up the aspect ratio (goes fullscreen rather than in a square). BUT, if I have ffdshow enabled, it messes this up every time.

no one answered my question earlier -- what are the effects of dscaler? what does it do, and what are the units of adjustment for it? I cannot tell what it is doing when I choose 'right half only'.

Hi:

I want to throw some numbers out here as a matter of interest. I am new to using ffdshow. I tried it earlier in my HTPC's life but abandoned it in favor of more pressing issues (i.e.; Girder, remote control, etc. and all things WAF). I have come back to using it now and am trying to learn quickly. I have not read all this thread (83 pages) but have reviewed the last 10 in earnest and have more or less listened to a lot of the ffdshow traffic over time.

What I am about to say may be of no revelation to most of you but going to put it out anyway. The focus of my info is centered around the cost of color space conversion by ffdshow.

From the reading I have done, it seems to be generally (my perception) recommended to output in YV12. I began to notice some anomalies in CPU consumption with and without the NVDVD-VPP (Video Post Processor) filter in the graph and did some investigating.

Note that like most codecs, NVDVD outputs YUY2 when stand-alone. However, with nVidia's NVDVD-VPP (Video Post Processor) inserted in the graph, VPP outputs in YV12. For my experiment, I used VPP to essentially change input to ffdshow from YUY2 to YV12.

The purpose was to observe the cost of (or lack of) color space conversion. For the numbers I am about to show, ffdshow is operating as a "null" filter: no resizing, no sharpening, no denoising, no nothing; all boxes unchecked. Also, all data was collected while in VMR9 mode.

The system is as follows:

Asus A7N8X MoBo
AMD-XP 2600+ (OC'd 10% ~ 2850+), FSB fundamental = 175 MHz
512 Corsair DC RAM (400 MHz rated)
Asus 9600XT
1 - Maxtor 120 GB @ 7200 PATA
2 - Maxtor 250 GB @ 7200 SATA Raid 0

Now for the CPU consumption info. These are average values that typically have +/- 5% excursions associated with them. Remember all values are in VMR9 mode and ffdshow in a "null" mode. Also note that DXVA does / will not work with VPP and / or ffdshow in the graph. The version of ffdshow is the latest non-SSE2 version and NVDVD is 3.0 Viewsonic release. Note that for the different color space permutations, I have forced output of ffdshow between YV12 (4:2:0) mode and YUY2 (4:2:2) mode.


NVDVD w/ DXVA ......................................... 20%
NVDVD w/o DXVA ....................................... 30%
NVDVD w/ VPP ........................................... 40%
NVDVD w/ ffdshow (null) .............................. 90% (YUY2 ---> YV12)
NVDVD w/ VPP and w/ ffdshow (null) .............. 40% (YV12 ---> YV12)
NVDVD w/ ffdshow (null) .............................. 45% (YUY2 ---> YUY2)
NVDVD w/ VPP and w/ ffdshow (null) .............. 50% (YV12 ---> YUY2)

Netting these numbers out (check my math please if you like) I come to the following conclusion:

ffdshow (null) YV12 ---> YV12 = 0%
ffdshow (null) YUY2 ---> YUY2 = 15%
ffdshow (null) YV12 ---> YUY2 = 10%
ffdshow (null) YUY2 ---> YV12 = 60%
EDIT: Please see Post #1671 for a clarification of the reason for high CPU usage for the this case (YUY2 ---> YV12). It turns out to be an issue with changing filter pararmeters within ZP but is of some importance that is worth knowing about since many probably do it this way since it is convenient and expeditious.

(Error for any of the above is probably around 5% (+/- 2.5%). I think the 2 middle numbers (10% and 15%) are more nearly equal than 5% different).

The color space conversion from YUY2 to YV12 is a killer. With YUY2 input, the best choice is to force output of YUY2. This is contradictory to some of what I have read or thought to understand. If input is YV12, then output should be forced YV12 but is not a killer even if YUY2.

My next experiment is to determine the CPU consumption with YV12 ---> YV12 vs. YUY2 ---> YUY2 for resizing to see the net advantage of being in the smaller YV12 color space. I expect YV12 ---> YV12 will have a lower net CPU consumption. I ran out of time.

While VPP helps the overall consumption scenerio, it draws 10% on its own. At this time no statement is being made about PQ with regard to VPP in the graph or not; it was being used to obtain a YV12 input to ffdshow. After I look at the resizing consumption numbers, I suspect there will be a net gain with VPP in the graph (YV12 ---> YV12) which right now is about a wash without VPP and YUY2 ---> YUY2 with ffdshow in "null".

I have no experience yet with the new WinDVD 6 codec yet but understand there is a registery dWord line item of "YV12FIRST" that supposedly will force the codec to output YV12. This would put this codec at an advantage for use with ffdshow particularly for those marginal on horsepower for getting the resizing they might not otherwise get.

This info may not apply to earlier versions of ffdshow. Hope I have all disclaimers in here. If I have overlooked something really essential, please point it out. This was just some quick number gathering and tabulating. I did not watch CPU trend for much longer than about 10 - 15 seconds on each pass but is enough for rough work. If this is all well known, forgive the redundancy and boredom and long-*ss post

Regards,
DFA

EDIT: Based on my info, I am currently using NVDVD 3.0 with VPP and ffdshow (YV12 ---> YV12) and sizing to 1920 x 1080 (no other filters) at about 80-85% CPU and VMR9. I tried using AVISynth / Lanczos per one of Vlad's techniques to down-res to 1280 x 720 after up-resing but can't quite make it. Need about 10% more horses. I think I can get it from my CPU but don't think the Lanczos down-res vs. the video cards HW bicubic down-res is worth the push. I might get it alone with WinDVD codec in YV12 mode but again, small PQ gain from my brief observation.


EDIT: Changed all occurences of wrongly spelled "YUV2" to "YUY2". I think I got them all.

Assuming that your methodology is correct, this is really helpful stuff, DFA. Thanks!

Interesting stuff DFA, but there's something fishy going on somewhere.

I use TT with ffdshow and based on your analysis it suggests that I should get 60% increase just by adding ffdshow, since TT (I.e. Sonic decoders) outputs in YUY2 and I output in YV12, which I do not. I'm at about 75% total utilization with both Gradual Denoise and Lanczos resize to 1440x960 compared to ~30% (going from memory for all these numbers) with just DXVA off.

Now, there was some earlier discussion that throwing in a gradual denoise, even at a very low level, somehow aligned the data structures or did the CSC quickly so that the net load on ffdshow was reduced. This may only be true for YUY2 -> YV12 and it's still a mystery, as far as I know, as to where the CPU hog is. This CPU reduction phenomena has been reported by others too.

If I get time tonight I will try something similar but since you're in experimentation mode I would be curious to see what you get if instead of a null ffdshow you use just gradual denoise with a strength of 5.

salsbst:

Thanks and I hope my numbers and assumptions are good. What I can say for sure is that using NVDVD stand alone (YUY2 input to ffdshow) and forcing YV12 output from ffdshow, it was all I could do to get an up-res of 1280 x 720. No way could I get 1920 x 1080 up-res until I matched input to output so I know there is something to it all even if I have got the reasoning wrong.

DFA

I thought TT outputted at YV12. I very well might be wrong. Love seeing stuff liek this though. One question, in NVDVD how do you enable VMR9?

TT outputs YV12 if you use DXVA but if you are in S/W mode it outputs YUY2.

Jvincent:

Yes, there is some things that don't quite make sense since, for example, under the adverse condition (YUY2 ---> YV12) of already being at 90% in "null" I can still bring in resizing to 1280 x 720 with little or no more CPU consumption until I start to go higher in resize. That does not quite compute. So there is probably more to it than I can extrapolate. It may also be something not quite right with the ffdshow version I am using. Not sure about any of it except that my eyes are still kinda good and this is what I get and what happens with my rig. Andy could probably give a better explanation.

For that matter, not sure why in "null" YUY2 ---> YUY2 is not "0%" since you would think this would also be just a pass-through like YV12 ---> YV12 which is "0%".

Also, guess I got all my "YUY2" wrong and put them as "YUV2". I will edit previous posts for sake of correctness. Still learning my "fourcc" stuff.


DFA

Ahh, thanks jv... I knew I'd seen it somewhere ;). Still curious how to force NVDVD into VMR9 mode.

madpoet:

NVDVD codec has an interface and tray icon when it is active. You can bring it up from the tray when active and set it for VMR9 only. Also, when bringing up the ffdshow dialog from the tray when running, it states the mode and what filters are in the graph at the very top of the dialog box. VMR9 was confirmed in all modes of test. I assume the results would be about the same in overlay, just stepped down in magnitude (note the use of the word "assume").

DFA

Great, thanks. Never thought to check the codec.

Just another quick couple comments.

With regard to my little test, I unchecked all output boxes except the fourcc type that was intended to be used. By passing your mouse cursor over the ffdshow icon in the tray, it will popup the input type, i.e. YV12, and the output type color space that is in use and should be done so for confirmation of what is acutually being used. Especially if trying to duplicate what I have found. The revision of ffdshow may also have everything to do with it as well. Take nothing for granted.

I have read a lot about Vlad's experiences and am his "padiwan" (sp). I also have Samsung DLP like Vlad that has good noise filter so see no need to employ ffdshow's as Vlad has also said. I also think Vlad is right in saying you can get the sharpening characteristics you want by manipulating the up-resing and down-resing. I don't have the horsepower to do a lot of the upsize and downsize manipulation but have quickly concluded this is where the lion share of ffdshow benefit comes from (resize) and have gone for as much as I can as opposed to bringing in other filters.

I know others just use resize as well.

My 0.02

DFA

Originally posted by DFA
NVDVD w/ ffdshow (null) .............................. 90% (YUY2 ---> YV12)
NVDVD w/ ffdshow (null) .............................. 45% (YUY2 ---> YUY2)
That's very odd, my CPU utilization is nearly identical when going from YUY2 input (WinDVD 6 filter) to either YV12 or YUY2 output, with the SSE2 optimized version.
Originally posted by DFA
I have no experience yet with the new WinDVD 6 codec yet but understand there is a registery dWord line item of "YV12FIRST" that supposedly will force the codec to output YV12.
It doesn't.

nm88:

Too bad on the WinDVD "YV12FIRST" thing. Not sure if it should be placed in the HKLM or HKCU part of the reg or if it is case sensitive.

No correlation for you then on CPU consumption. I was looking for a "March" release of ffdshow because Vlad had stated that this consumed less CPU for resize than others he had tried. I would also like to see if what I have found duplicates with some other release.

I am giving thought to what else might be in play that I have overlooked but nothing comes to mind.

What DOES come to mind is one time on the Sage Forum I commented that SOMETIMES during a live TV commericial, I could not bring up my browser from the tray like Sage was in an "Always on Top" mode. As soon as the commercial ended or was in between commercials, I could bring up the browser. As far as I know, no one else was ever able to duplicate it. It was an off-the-cuff comment but sparked some interest that I now wish it had not gotten. I still stand by that one or will at worst give in to an incredible statistical phenomena. I just hope I'm not stretching or lossing what credibility I have left with yet another unduplicatable observation!!!

When I get home, I am going to recheck the "90%" data point (YUY2 ---> YV12) and make sure all parameters are as they should be. Also, will publish exact release date of ffdshow in use.

I wonder if for AMD XP if 3D-NOW or SSE is being used. ffdshow has the capability to use either according to the grayed out check boxes.

DFA

Madpoet, you must do an image of the system drive first if you want to try to disable HT. I have the same GA-settigns in the BIOS as you, as far as I can remember. Windows may even have to be reinstalled or maybe installed on top of the old version since it uses another HAL (Hardware Abstraction Layer, but I'm sure it was named in reverse, they decided they wanted to use the acronym HAL (2001 A Space Idiocy - note that HAL + 1 letter each is IBM) and just found something that it could mean!). Don't try if you haven't got Norton Ghost or PowerQuest Drive Image.

I turned off HT last week... Without reinstaling anythting, and it works just fine...
Maybe I've been lucky, but I just turned HT off from BIOS and now it shows as one thread in task manager and ffdshow seems to run smoother...

BTW I am NOT saying that it is SAFE to disable HT without doing a backup... just saying what I did... (and I did it that way because I never thought of the possibility of data corruption)

Turning off HT on a machine that was installed with it on is perfectly safe (still MP HAL) but performance won't be optimal. Even turning it on on an install where it was disabled should be safe but it won't get used. I don't think there are any conditions where XP will use 2 processors when it doesn't know how to.

-Pat

Originally posted by DFA
nm88:
I wonder if for AMD XP if 3D-NOW or SSE is being used. ffdshow has the capability to use either according to the grayed out check boxes.

DFA

Little info even if SSE is showed in the detection box there is no function wich use SSE code, since SSE is useless for ffdshow :)
There is also no real 3D-Now code in ffdshow (we also dont need 3dnow), since we just work with integers and so 3dnow == mmx2. So dont be confused by the feature settings in the ffdshow cpu tab.

Its like:
mmx1 = integer instructions in 64bit registers
mmx1 + 12 new instructions = mmx2
mmx2 + floating point instruction's = 3d-now1

3dnow in 128bit registers = sse1
sse1 == 3dnow ext (or 3dnow pro)

sse1 + 128bit integer instructions = sse2
sse2 + 8 new instructions = sse3

So if u dont work with floats u dont need 3dnow or sse1.

DFA: Thank you for your analysis. It is great to see someone spend time to try to quantify and in a well structured way... keep it coming!

Following DFA's lead, here's some data for TT (Sonic) with ffdshow. I left out the DXVA mode since it's not really relevant to our discussion.

These numbers are on a 2.5G P4 with PC2100 RAM and using the June 16, SSE2 optimised build. I'm also running reclock and DVDidle outputting at 1440x960 via the component output of an AIW 9700 Pro if that matters.


TT w/o DXVA, no ffdshow ........................................... 20%
TT w/ ffdshow (null) ................................................... 80% (YUY2 ---> YV12)
TT w/ ffdshow (small Gr Denoise) .............................. 45% (YUY2 ---> YV12)
TT w/ ffdshow(Gr Denois + resize 1440x960 lanczos) 75% (YUY2 ---> YV12)

*TT w/ ffdshow (null) ................................................... 80% (YUY2 ---> YUY2)
*TT w/ ffdshow (small Gr Denoise) .............................. 45% (YUY2 ---> YUY2)

Not sure what exactly is going on but it looks like if you bypass any ffdshow processing you hit some nasty, slow piece of code. I need to double check the YUY2 to YUY2 cases since I didn't restart the player when I switched the output colourspace but I can't right now since the kiddies commandeered the TV/HTPC.

Originally posted by jvincent

Not sure what exactly is going on but it looks like if you bypass any ffdshow processing you hit some nasty, slow piece of code. I need to double check the YUY2 to YUY2 cases since I didn't restart the player when I switched the output colourspace but I can't right now since the kiddies commandeered the TV/HTPC.

mhh i just profiled the code with Windvd5 + abstract + ffdshow (null) (overlay mixer and zoomplayer)

YUY2 (uncompressed) from the dvd to -> yv12 output (no filters)

ivivideo.ax (15%)
ffdshow (5%)
zoomplayer (7%)
ac3filter (3%)

YUY2 (uncompressed) from the dvd to -> YUY2 output (no filters)
ivivideo.ax (13%)
zoomplayer (7%)
ac3filter (3%)
ffdshow (1%)

So for me ffdshow without a filter and with same input than output YUY2-YUY2 i get +1% cpu load from ffdshow.

In general ffdshow will convert every input format (if a filter is enabled) internal to yv12 and than convert yv12 to the output format.

So for me all looks oki?

PS: cant check for the NVDVD3 since i dont have this filter, the cyberlink and elecard seems to show the same results on my AMD64 system.

Jvincent and Andy:

Jvincent: Interesting that you were able to get something similar but I have some important info to add. Were you making the changes from within ZP while ZP was just stopped?

As the day wore on, I was thinking, "No way a color space down conversion could consume 60% CPU. Something has got to be wrong." I finally dropped what I was doing and went home to recheck things.

It seems that what I said before is true but there is some sort of problem with Zoomplayer and the codec and / or ffdshow.

I am using ZP ver. 4.00 RC2. The ffdshow.ax says it was created May 20, 2004.

Here is the way I was doing my testing: I would "Stop" ZP, open the ZP config panel and highlight the ffdshow filter. Then push the "C" (configure" button to bring up the ffdshow configuration applet and make my changes along with removing and adding VPP to the filter graph. It seems this is where things go wrong.

If I configure ffdshow (null) for the YUY2 ---> YV12 case when ZP is completely closed then open ZP with ffdshow and ZP pre-configured, I ONLY get a total 40% CPU consumption. On the other hand, doing it from within ZP, even though stopped, making changes and restarting for the YUY2 ---> YV12 case, I can get the total 90% consumption as originally stated. I can repeat it with the proper sequence of events. So even though it does occur, it is a bug of some kind it seems.

Nevertheless, may be worth knowing about because I also found I could drive my 1920 x 1080 resize setting with VPP and ffdshow from 80 - 85% to 90-95% by manipulating things from within ZP.

The saga goes on even further. After understanding the issue and writing it off to a ZP/ffdshow/NVDVD problem, I tried to bring Sage out of sleep. I had not tried Sage since this early morning's testing so was not aware there was a problem waiting. Sage also uses the NVDVD codec in VMR9 and has for a long time with no issues. Sage error-ed with not being able to connect to the codec. A complete cylcling of Sage did not help. It would run in overlay but not VMR. Ultimately, a reboot and changing it from DXVA MPEG mode 2 to default brought it back. It has been in mode 2 for a long time w/o difficulty.

Still not the end of things. I went back and ran ZP again after getting Sage straightened out and found my CPU usage for YV12 ---> YV12 (VPP in the graph) and with a Lanczos resize of 1920 x 1080 had lowered from the previous and consistent value of 80 - 85% to 75 - 80%.

So, one or more things are not playing nice with each other. Not sure who the culprit is either. Just that I know now how to make it happen and how to avoid it. It seems it can be done with TT as well.

My problem now is what to do about my not entirely accurate first post on the this subject? Should I just amend the top of it directing to this post # for an update or what? That's what you get when you shoot from the hip.

Andy:
The profiling shows just who is using what. How do you manage that? Were you just moitoring the "processes" in Task manager?

DFA

Originally posted by AndyIEG
In general ffdshow will convert every input format (if a filter is enabled) internal to yv12 and than convert yv12 to the output format.
Why would ffdshow purposely compress a YUY2 signal to YV12, if the output isn't YV12? Isn't there a way to maintain all of the information in the original signal?

Gosh, here I was wondering why DVD decoders whould output YUY2 when all of the information on the DVD fits into YV12. My conclusion was that I misunderstood the DVD spec, but that it should be changed to YV12 so that I can run ffdshow on my HTPC.

Originally posted by DFA
Jvincent and Andy:

Jvincent: Interesting that you were able to get something similar but I have some important info to add. Were you making the changes from within ZP while ZP was just stopped?



I don't use Zoom. I was trying this with TT. I was making the changes without quitting TT. This is valid for all of the changes, with perhaps the exception of the colourspace change.

Kids are in bed so I just double checked and I get the following:

TT w/ ffdshow (null) ................................................... 25% (YUY2 ---> YUY2)
TT w/ ffdshow (small Gr Denoise) .............................. 30%% (YUY2 ---> YUY2)

So it looks like if you switch output colourspaces you need to restart the player AND there is something funny in the colourspace conversion routine when no filters are selected.

Jvincent:

I knew that but forgot (you use TT).

Try quiting TT for the ffdshow (null) YUY2 ---> YV12 case and see if CPU is nearer 40% instead of 80%.

I also am convinced something is amiss but think it convoluted.

EDIT: Never mind. I guess you already stated that was your procedure to cycle TT for color space change.

DFA

Originally posted by DFA
Andy:
The profiling shows just who is using what. How do you manage that? Were you just moitoring the "processes" in Task manager?

DFA

umm? Just shows who is using what, thats the same u tryed? It shows that ffdshow raise the cpu usage by 1% without any filters.

Im using a profiling tool wich make use of the AMD/P4 internal cycle counters and i can exact see how many cycles/cpu usage every process has. The data the taskmanager provide are useless and very inaccurate.

Its the same tool i use to test my code and spot where the problems are or wich function takes what cycles. I can pinpoint the cpu/cycle usage down to a single asm instruction and also get an idea how the pipeline is filled.

There are many profiling tools, i use the AMD codeanalyst and intels Vtune.

Originally posted by nm88
Why would ffdshow purposely compress a YUY2 signal to YV12, if the output isn't YV12? Isn't there a way to maintain all of the information in the original signal?

Its cause the internal filters need yv12 format, im not 100% sure but i dont think u can see a diff. in image quality from yuy2 to a good yv12 conversion.

Originally posted by DFA
Jvincent:

I knew that but forgot (you use TT).

Try quiting TT for the ffdshow (null) YUY2 ---> YV12 case and see if CPU is nearer 40% instead of 80%.

I also am convinced something is amiss but think it convoluted.

EDIT: Never mind. I guess you already stated that was your procedure to cycle TT for color space change.

DFA

Actually I didn't try quitting TT before the "Null" YUY --> YV12 and when I tried that it was around 30%, which makes more sense.

Just couple of comments. I did use March 25th version of FFDShow for a while as April and May builts did not work well with Forceware Multimedia (they did as long as you did not oppened configuration box for post processing, FFDShow would vanish without the trace...forewer). I tried one of Andy's earlier builds and got some stuttering, but now I am using his June 16th SSE2 built without any problems.

As far as colorspace conversion, I realy like YV12 input and output. That is the only combination that really resolves DVE chroma resolution pattern perfectly. Lately I have been lazy so all I have on my HTPC is Sonic and Nvidia decoders. I tried Elecard, verified it looks great, but overall still like Forceware better due to a bit smoother payback, and full support (subtitles, macrovision, etc.). So I am not to woried about colorspace conversion righ now as YV12 to YV12 works perfect. But it is interesing to see how things get messed up realy easy with HTPC.

I did test FFDshow resize to 2560x1440 (twice my output resolution of 1280x720, thanks Andy and Milan for fixing this) and found out that my card has issues with this input signal. I did see the image, but there was also a big mess with some horizontal green lines on sides of the screen. From what I saw in the middle part (which was clear), I could not realy see any meaningfull benefit to additional upsize compared to 1920x1080 upsize. This is kind of bad news, but I will get back to it once new SSE2 optimized version comes out.

As some of you might have noticed, I am playing with GPU 3D settings right now trying to get some "CPU free" benefits from AA and AF algorithms. It looks like it kind of works. I say "kind" of because it is all bloody beta mess and activating it is like a ritual, one wrong step and you are doomed.

Currently i have a celeron 2.8 ghz and 512 ram (winxp pro + sp1) and TT 1.5 . Do you think i can handle (my cpu i mean) ffdshow with scaling , sharpness and denoise filters ? I can't go higher (p4 3 ghz) because my motherboard is old and accepts only 433 bus cpu.

Currently i have a celeron 2.8 ghz and 512 ram (winxp pro + sp1) and TT 1.5 . Do you think i can handle (my cpu i mean) ffdshow with scaling , sharpness and denoise filters ? I can't go higher (p4 3 ghz) because my motherboard is old and accepts only 433 bus cpu.

You can certainly do SOME resizing. How much is going to be trial and error for you. But I would imagine you can double resize (1440x960).

Originally posted by vpopovic

I did test FFDshow resize to 2560x1440 (twice my output resolution of 1280x720, thanks Andy and Milan for fixing this) and found out that my card has issues with this input signal. I did see the image, but there was also a big mess with some horizontal green lines on sides of the screen. From what I saw in the middle part (which was clear), I could not realy see any meaningfull benefit to additional upsize compared to 1920x1080 upsize.

milan just lifted the restriction in the ffdshow pannel, but i forgot to also raise the resizer buffers :( Thats why u see this crap at resolutions higher than 2000. In the next version this is fixed.

PS: for resolutions higher than 2000x2000 u need VMR7/9 since overlay mixer wont connect.

I read through the last 10 pages and I did a search on Denoise3d but I just can’t seem to find where you set it up in the configuration program? Can someone help a newbie out with this?

Under the Blur/NR menu. ;)

Thanks sorry for my ignorance.

Denoise3D and Owen are my heroes! My X1 experience is so much better now.

I run 1440x960 without problems, but when I change it to 1920x1280 it works fine on some movies, on others the video stutters for a few seconds periodically. Is there a setting between these two? 17 something x something?

You can try 2.5x instead of 3x.

-MP

I just get worried if FFDShow thread is not on the first page. Are we running out of ideas, or people just think FFDShow is not worth its CPU toll? I still find it the best software tool around, by far...and if I was starting my HTPC journey, or wanted to take it to a higher level, I'd have lots of questions.

I have the following HTPC setup:

Samsung HLN467W DLP
AMD64 3200+ 2.0Ghz (OCed to 2.14Ghz)
ASUS K8V SE Deluxe MB
Sapphire Radeon 9800Pro 128MB flashed to 9800XT 414/365Mhz
Mushkin PC3500 Memory 1GB
Omega 2.5.5.1 Drivers
Zoomplayer Pro 4.00 Final (trial version)
FFDSHOW 20040616 (SSE2)
Windows XP Pro SP1 updated
WinDVD5

I have Zoomplayer set up to use WinDVD5 for the audio and video decoder. The audio renderer is Default DirectSound device and the video renderer is set to Overlay Mixer. The additional filters I have are DMO_V InterVideo Abstract followed by FFDSHOW Video Processor. I have modified the registry based on other posts to set Abstract intesity to 0.

If I try simply to enable Lanczos resize to 1280x720 in FFDShow my CPU utilization shows ~100% in Task Manager. When this happens the DVD just becomes jumpy because it seems frames are being dropped.

Based on other posts in this forum it seems that others are achieving much more than this with equal or lesser systems.

Can anyone give me suggestion on how I can troubleshoot this or what I may be doing wrong? Or is my system running as expected? I would like to also note that if I take FFDSHOW and DMO_V InterVideo Abstract completely out of additional filters in Zoomplayer I'm still getting about 30% cpu hit.

Any help appreciated. Thanks....

-Rahul

For reasons that do not seem to be well understood, you need to enable some filter before (above) resize, or your CPU utilization will be 100%. This is why all the suggested settings start with blur/NR or a levels adjustment even if they are effectively nulled out.

Add a filter before your resize and you should see the utilization come back down to reasonable levels.

Wondering what codecs will output YV12.

Elecard build 2510 does (ie if one sets FFDShow input YV12 it will work and input shows as such)

WinDVD5 with the abstract.dmo does not

Cyberlink PDVD5 does

Haven't got any others to test.

WinDVD is stupid :( hehe Im currently using it for DVD (elecard for file) as elecard does not have menu select support in build 2510, and newer supporting versions have dirtyChroma boy! :)

How much worse is Cyberlink PDVD5 vs WinDVD codec?

Wondering what codecs will output YV12.

Elecard build 2510 does (ie if one sets FFDShow input YV12 it will work and input shows as such)

WinDVD5 with the abstract.dmo does not

Cyberlink PDVD5 does

Haven't got any others to test.

WinDVD is stupid :( hehe Im currently using it for DVD (elecard for file) as elecard does not have menu select support in build 2510, and newer supporting versions have dirtyChroma boy! :)

How much worse is Cyberlink PDVD5 vs WinDVD codec?

Something's not right Spoonfed. I use Windvd5 and YV12 no problems.

CODEC section of FFDShow set to YV12 ONLY? hmmmm
Gives me errors.

I'm using 5.3.5.22 on an Athlon (I've heard of problems with certain Windvd versions and P4's).

Athlon also here
in WinDVD says 5.1 B05.005

Version of the filter file is 5.1.5.5

So Perhaps there are changes. I wonder if the newer WinDVD 6 would work, hmmmmm.

If you install the "demo" version, will it still let you use the codec after expires?

johnbrisbin,

Even if I have another filter before Resize such as Blur/NR enabled but effectively nulled I'm still seeing 100% CPU utilization. However, if I run FFDSHOW with nothing enabled and then start enabling filters while the DVD is playing via the FFDSHOW system tray icon my CPU utilization will not go to 100%. I'm not sure why that makes a difference...

Originally posted by Spoonfed
CODEC section of FFDShow set to YV12 ONLY? hmmmm
Gives me errors.

If you are using ZP with ffdshow, windvd video decoder and the abstract filter you will get an error when chosing only YV12 as input (need YUY2). If you remove the abstract filter it will work.

So is one of the ultimate scenarios for running ZP with FFDShow as follows?

ZP:
Elecard 2510 with VMR9 Windowed
WinDVD audio codec
Reclock

FFDShow:
1. Denoise3D - 0.5, 1, 5, HQ
2. Lanczos resize @ 1440x960 with Luma=1.1 & Chroma=0
3. Overlay (unchecked), but in Output area of Overlay, YV12 as only option checked.

Hey all,

for some reason I'm getting a much cleaner picture with WinDVD5 and ffdshow than with WinDVD6 and ffdshow. Using the exact same settings in zoomplayer and ffdshow. Anyone, notice similar issues? In particular, it seems to me that the denoise3d for some reason works much better with the winDVD5 decoder? I have no idea why that would be the case? Any idaes?

How do I turn the ffdshow filter on for DVDs in Zoom Player 4. I have been going into the video codec section in Zoom player and adding the ffdshow filter to the MPG2 codec. However, someone else mentioned that they added the filter in the DVD section where it says additional filters. I tried that and it gave me an error. Where do you guys set the filter up for DVDs?

That's where I do it... in the additional filters.

mpgxsvcd,

Sounds like you don't have FFDShow configured properly. Make sure you have Raw Video set to 'all supported' under codecs in FFDShow configuration.

-AYB

pbpatel98:

I use resize to 1920 x 1080 with a AMD 2600+ OC'd 10%. CPU consumption is average 80%. So, yes something is wrong because you should do better.

I am using NVDVD 3.0 codec with VPP (VPP outputs YV12) for YV12 in and YV12 out. Save some overhead in ffdshow since according to Andy, ffdshow will down convert to YV12 internally anway. I found this to cost me about 10% CPU in ffdshow so why not give it to the VPP which also consumes about 10% CPU and seems to give me a little better PQ. (that was opinion)

Back to your problem. A few pages back I had made an issue about color space and CPU consumption. After some back and forth and going around, I found that making changes to ffdshow should be done with the player (ZP) completely closed. Make changes in ffdshow applet then close applet before restarting player. I found that when I was just stopping ZP and bringing up the ffdshow applet within the ZP application, I was getting high CPU consumption (+90%) even with ffdshow configured null. This also seemed to stick after the fact once done this way. I also found I could get near 100% CPU usage with my resize (normally 80%) when tinkering with ffdshow settings from within ZP when ZP was only stopped but not closed.

Andy probably needs to take a look at this. It was duplicated by jvincint with TT and Sonic (80% with null configuration but configured w/o completing closing TT; only stopping and 30% when closing TT then restarting TT for the same ffdshow configuration) so is probably an ffdshow quirk.

This may or may not be your problem but be aware of it. I can duplicate it at will and also now know how to get around it or kill it if it happens.

EDIT: Also pay no attention to CPU consumption at trailer stuff in beginning of DVD. Be sure you are into the movie (film) when looking at CPU consumption.

DFA

Originally posted by AllYourBass
mpgxsvcd,

Sounds like you don't have FFDShow configured properly. Make sure you have Raw Video set to 'all supported' under codecs in FFDShow configuration.

-AYB

Do I need to mess with the Raw-only configuration utility that comes with the latest version of ffdshow also or just the configuration utility?

Originally posted by JoeFigueiredo
So is one of the ultimate scenarios for running ZP with FFDShow as follows?

ZP:
Elecard 2510 with VMR9 Windowed
WinDVD audio codec
Reclock

FFDShow:
1. Denoise3D - 0.5, 1, 5, HQ
2. Lanczos resize @ 1440x960 with Luma=1.1 & Chroma=0
3. Overlay (unchecked), but in Output area of Overlay, YV12 as only option checked.


This setup is definitely excellent, but I will make a couple of comments.

First, the Elecard decoder (2510) has known aspect ratio issues for NTSC users, but works perfectly for PAL. Also sub support is missing, so menus will be a problem. But, and this is only thing that matters to me, image quality is second to none.

Second, VMR9 output can look a little soft at 1440x960 or less.
The higher you resize, the sharper VMR9 gets, so for users how are limited in resize ability, Overlay will be noticeably sharper.
For VMR9 output, I prefer to use VERY high resize settings to avoid the softness. I use 1920x1728 for PAL. 1920x1440 or 1440x1440 would be good for NTSC users.
I understand that these resize settings may sound ridiculous for use with VMR9 and will be imposable for many, but if you have the power, I highly recommend it.

As for resize type, I prefer Bicubic over Lanczos.
Bicubic looks smoother, more natural and less digital on my display.
This is probably because my display is very sharp and I don’t need any sharpening in FFDShow. Even Luma 0.5 is over the top on my display.
YMMV, so just try both Lanczos and Bicubic and make your own choice.
Resize sharpen settings vary greatly depending on display and personal preference, so don’t just use some one else’s settings. Adjust Luma sharpness to your taste, but don’t sharpen Chroma as it will only increase noise.
Remember, less is more with sharpening. If you go for a dramatic increase in sharpness, it may be impressive at first, but you will probably find that over time you will come to dislike it.

Regards to all,

Owen
____________________________
The FFDShow resize sharpen dude.

DFA,

I read the post about completely stopping ZP and then performing changes in FFDSHOW and then restarting ZP. Unfortunately, that doesn't seem to help me too much either. What is the FSB of your system? Could it be that since my Athlon64 is running at only 200Mhz on the FSB that it's not fast enough for FFDSHOW?

mpgxsvcd:

I use "Raw only configuration" but does not seem to be a necessity. "Raw only configuration" appears to just do away with the listing for the large array of input codec / decoding choices and just displays the "raw video" item. All the rest should be disabled anyway if not using "Raw video configuration" so is same result.

I tried both ways and only noticed the change in applet appearance but no difference in performance.

DFA

pbpatel98:

That can't be your problem because I have one of the earlier A7N8X MoBo's which is 166 MHz FSB fundamental (x2 for DDR RAM and x2 again for DC mode) OC'd to 175 MHz. You actually have an effective 800 MHz FSB RAM speed assuming you have DC (Dual Channel) working.

Are you sure you are using a "film" DVD and not video? You seem to be suffering from the "unexplained CPU consumption" thing but what you are doing gets me around it so don't know what to say.

You have the horsepower for sure. Check the "check boxes" in the resize dialog to make sure you have everything as it should be.

I'm a newbie at this myself so know just enough to be dangerous.

EDIT: Are you using VMR9 or overlay? Start with overlay as a test; it is easier in all regards. I am using VMR9 but if I were having problems would roll back to overlay for simplicity and problem isolation.

DFA

DFA,

What would be a "film" DVD as opposed to video. Lately I have been using the movie S.W.A.T. to test with because it was the first one that popped out at me in my collection.

I guess I'll have to keep scouring the forum until I can come up with a result. Thanks for your help though...

pbpatel98:

I don't know how you know in advance of using it (looking at the DVD) and am ignorant about S.W.A.T. I have wanted to ask the question myself to know how one knows by looking at the disk or box.

One way to tell is frame rate. Video frame rate will be 30 fps (29.9xy to be more precise but do not recall x and y). Film will have frame rate of 24 fps. Video material behaves entirely different than film and is frequently interlaced as well. One thing for sure, resize horsepower goes up alot for video material. I have assumed that is due to the 25% higher frame rate alone. Video based trailers and previews peg my CPU at 100%.

Others can clarify this better and I would not mind hearing more on the subject myself (and corrected if I have mis-stated) although probably discussed ad nauseum through out many threads.

DFA

This article helped me a ton to comprehend whats going on with MPEg decoders/encoders / ffdshow and other such filters - and it explains the CHROMA BUG or CUE Chroma Upsampling Error.

http://www.hometheaterhifi.com/volume_8_2/dvd-benchmark-special-report-chroma-bug-4-2001.html

To my newb logic ... DVD Source material Progressive Scan or Interlace will suck up cpu, based upon the decoder you choose. [meaning if it incorrectly decodes EVERY source it sees via the Interlace algorithm, whether the source material is flagged Progressive Scan or not!] I gotta think Progressive Scan Source material is the least taxing on the CPU. Thats just a newb guess though.

SOCAL-HTPC:

Yes, that should be required reading for all HTPC enthusiasts. I read it shortly after getting involved in HTPC and re-read it recently and understood more the second time.

But there is a fundamental frame rate difference between film and video and that is where I believe the CPU consumption difference lies.

DFA

How do I know if ffdshow is actually being applied to the DVD that I am watching?

Originally posted by mpgxsvcd
How do I know if ffdshow is actually being applied to the DVD that I am watching?

You could simply check the OSD in ffdshow. If it shows up you know ffdshow is active.
_____
Axel

Originally posted by DFA
pbpatel98:

One way to tell is frame rate. Video frame rate will be 30 fps (29.9xy to be more precise but do not recall x and y). Film will have frame rate of 24 fps. Video material behaves entirely different than film and is frequently interlaced as well. One thing for sure, resize horsepower goes up alot for video material. I have assumed that is due to the 25% higher frame rate alone. Video based trailers and previews peg my CPU at 100%.

Others can clarify this better and I would not mind hearing more on the subject myself (and corrected if I have mis-stated) although probably discussed ad nauseum through out many threads.

DFA

From my Divx/xvid ripping days, Video/Film was mainly described as interlaced/progressive (not interlaced). Keep in mind that I really do not have an extact grasp of the theory but knew enough to help ripping...

NTSC generally is interlaced at 29.9 fps. The original film is 23.9 fps, when interlaced it adds the exta frames in order for most NTSC displays to work.

When ripping we would de-interlace, or remove the extra frames used in interlaced DVDs to reduce the frames being processed and thus save space...

Its good this came up because I was wondering some of the same things as you DFA.

I know ffdshow has a deinterlace filter, but was not sure if the same thoery of deinterlacing would hold true for ffdshow, it would make sense, less frames less work.

I also am not sure if desktop resolution plays a part, whether it is interlaced or progressive...

Owen,

I did some playing last night... it seems that now I have the Radeon 9600 instead of the 9000 I can see noise I couldn't with the 9000... picture res doesn't seem different though. However, now a little denoise 3D is of advantage...

Does Denoise3D still have the colour problem ?
It seems to me that it still does and this is a real problem (and don't tell me to change my color settings, a denoise filter should not touch the colour :)).

No one replied earlier, and i'm having a hard time seeing a recent (not from 2 years ago in this thread) explanation of the advantages of the dscaler section of ffdshow.

I'm running mine @ 100 units, and i cannot tell what it is doing. Am I just better off not running it and increasing my resolution instead [since more CPU power will be opened up]

When i run it w/ 'right half only' I cannot tell at all what is happening.

Originally posted by BangoO
Does Denoise3D still have the colour problem ?
It seems to me that it still does and this is a real problem (and don't tell me to change my color settings, a denoise filter should not touch the colour :)).

The latest version should not change the colors, i fixed this and tested the code against the old c++ version and the SSE2 values are 100% identical to the c++ values.

Just enable denoise3d in the sse2 version take a screenshot and compare it against an older 2003 version bit by bit. I will double check the output again if i release a new SSE2 build, just to be sure.

Thx Andy.
I use the latest and it looked like the problem was still there... So I did some snapshots and the colours seem to be the same with or without using Denoise3D.

Originally posted by Axel
You could simply check the OSD in ffdshow. If it shows up you know ffdshow is active.
_____
Axel

Ok I must be doing something wrong. I don’t even know what OSD stands for? I did a search for it and it appears to be a statistics box. When should I see it with zoom player 4.0 pro? How do I enable it? What does it say if it is not working but it is enabled. Sorry about the questions but I am new to ffdshow and I really couldn’t find the answers using the search. Any help is appreciated.

Originally posted by BangoO
Thx Andy.
I use the latest and it looked like the problem was still there... So I did some snapshots and the colours seem to be the same with or without using Denoise3D.

Maybe u noticed the "green" shift bug wich is still in, but better masked than the 2003 versions. I try my best to mask both in the next version, but its a crazy problem, without an easy and perfect solution.

OSD = on screen display.

ffdshow's happens to display some statistics on dropped frames, I believe. If you don't see, it isn't enabled or isn't working properly.

Originally posted by mpgxsvcd
Ok I must be doing something wrong. I don’t even know what OSD stands for? I did a search for it and it appears to be a statistics box. When should I see it with zoom player 4.0 pro? How do I enable it? What does it say if it is not working but it is enabled. Sorry about the questions but I am new to ffdshow and I really couldn’t find the answers using the search. Any help is appreciated.

And to answer the other parts of the question, it is a checkbox in the FFDShow config (up at the top) and it'll either show up in the top-left of the screen or it won't. If it does then ffdshow is working, if it doesn't then ffdshow isn't processing the frames.

-Pat

I've been happy using Zoom Player and ffdshow for a few weeks now, getting good at the tweaking, etc.

Despite the excellent picture, I've noticed a "comb" effect occasionally. I don't think I have the de-interlacer turned on, probably because I never saw it mentioned specifically in any of the recommended settings (although people have been talking about dScaler).

I have a Samsung HLN4365W, and have ffdshow output at 1280x720 (resolution the desktop is set to), with picture adjustments (for color/brightness), some blur/noise and sharpening if needed, etc.

Should I be turning the de-interlace function on? What about a dScaler? And if I should have it on, do I have it before or after the resize (I would think before, so it does less work...)?


TIA

Originally posted by AndyIEG
Maybe u noticed the "green" shift bug wich is still in, but better masked than the 2003 versions. I try my best to mask both in the next version, but its a crazy problem, without an easy and perfect solution.
Maybe... but now that you told me it still exists, it bothers me :)
Why is there a green shift on a denoise filter (I guess Gradual Denoise does not have this problem) ?

I read this thread from page 60 on, skipping some of the messages which I did not understand much. I have some questions for Andy and Owen and other AMD experts are:

I was all ready to build with P4 2.8c P4P800e dlx board. But with the news of all the new stuff from Intel, I had second thoughts and had done a lot of reading on AMD64 and MSI K8N Neo Platinum nForce 3 250 Gb. My current thinking is that the AMD64 platform will give me some future proof since it is 64 bit. From reading this thread, it gave me the impression that AMD64 will do for SSE2 version and memory bandwidth does not matter very much. If I am not wrong, my question now is what is the minimum AMD64? Should I buy 3200+ or 3000+ (a $100 difference) ? It can be overclocked since mobo allows locking AGP/PCI. I want to be able to do Denoise3D, resize, sharpen. What cpu is good enough? If I want to do Resize first followed by denoise3D, what cpu should I get?

My second question is that Windows 64 extension will be out soon, should I wait for it and also wait for Andy's new 64 bit version? If Andy's 64 bit version is not ready for some time, will the SSE2 version run in Windows 64?

For memory, I plan to use 512 MB KingMax Hardcore PC4000 DDR500 (good quality memory for dual channl according to some reviews, but I'll use it for single channel). ATI Radeon 9600XT for video.

Thanks you for comments and help to get me started. I have been reading for the last 3 months, and hope to start to build and have some real fun soon.

You need more power ;). Seriously, if you want to resize before Denoise then it takes some oomph. If you are looking to futureproof anyway, get the most bang you can afford.

Thanks for the quick reply, Madpoet. 3400+ cost $300 more than 3000+ and $200 more than 3200+ (and then there are higher ones and FX series). I want to future proof, but have to draw the line. I guess I'll have to give up resize before denoise. Can 3000+ do the job for Denoise followed by resize or do I need the 3200+?

If you're OCing anyway, 3000+ will be ok.

e268,

I have the following HTPC setup and I can't get any resize above 1280x720:

Samsung HLN467W DLP
AMD64 3200+ 2.0Ghz (OCed to 2.14Ghz)
ASUS K8V SE Deluxe MB
Sapphire Radeon 9800Pro 128MB flashed to 9800XT 414/365Mhz
Mushkin PC3500 Memory 1GB
Omega 2.5.5.1 Drivers
Zoomplayer Pro 4.00 Final (trial version)
FFDSHOW 20040616 (SSE2)
Windows XP Pro SP1 updated
WinDVD5

I'm still convinced that something must be off w/ my setup because others are able to do more with the same config or less. I have read every relevant post I can find about ffdshow/zoompayer configuration with no luck.

If I were to do it again I would get an FX processor to take advantage of dual channel memory bandwith and make sure I get a MB that allows to me lock the AGP/PCI bus so that I can overclock more.

Originally posted by BangoO
Maybe... but now that you told me it still exists, it bothers me :)
Why is there a green shift on a denoise filter (I guess Gradual Denoise does not have this problem) ?

nah i mean the green shift bug is in the resizer, so u prolly noticed it in combination with the denoise3d.
All my test showed that the old bug is smacked and u dont get any color diffs. by using the denoise3d sse2 version. Feel free to test and if u find any color problems send me a message, but atm i get bit by bit the same result if i take a screenshot with the old and new code at the same frame.

@e268
Im still not sure if i can build a working AMD64 version, i have to rewrite some code and redo quite a lot asm code. The main problem is not the asm code. Its more a try and error work for the normal c++ code...
But i can say that if i had all the new 64bit stuff avaible for my actual work on the resizer the code would run a lot faster.

My advice for now is to get a P4 or AMD64 and if u can spare some bucks try to get a version with more CPU cache. Atm i have only infos on the AMD 64bit stuff, i have no clue what intel is up to.

Ok thx for the explanation.
I'll keep on using Denoise3D then :)
It's the 1st time I hear about a green shift bug in the resizer though... it's there on every resizer (lanczos, bicubic, etc...) ?

Originally posted by BangoO
Ok thx for the explanation.
I'll keep on using Denoise3D then :)
It's the 1st time I hear about a green shift bug in the resizer though... it's there on every resizer (lanczos, bicubic, etc...) ?

The green shift "can" be visible in every resize mode, but how more taps how more shift. So its most noticeable if u use Spline or Sinc since those have the highest tap deep. But the green shift is much better masked in the 2004 versions.

The horizontal lining bug is just visible if the Y resolution is not a direct multipler of the org. Y resolution and again more taps = more visible lining's.

Ok thx again :)
I use Lanczos at 1440*960 for NTSC and 1440*1152 for PAL so I guess it's ok...

I have a Athlon 2400+ moderately OC'd to a 2600. I get almost 100% CPU usage just using the NR. When I try the resize I get a constant 100%. I am using the latest version of FFDshow as well as ZP. I am using ZP with the powerdvd and teh sinemaster codecs. What is teh best version of FFDShow to use, and how do I reduce CPU usage.

Originally posted by sknyfs
I have a Athlon 2400+ moderately OC'd to a 2600. I get almost 100% CPU usage just using the NR. When I try the resize I get a constant 100%. I am using the latest version of FFDshow as well as ZP. I am using ZP with the powerdvd and teh sinemaster codecs. What is teh best version of FFDShow to use, and how do I reduce CPU usage.

Try the 20021213.exe version.

sknyfs,

I would say forget about those old ffdshow versions, Andy's newest with optimized denoise3d are the best :) Because of that and changing back to bicubic resizing (thx to Owen) I can finally use my XP 2600+ Barton for the settings I wanted: denoise3d followed by 1440x960 resize + luma sharpen in VMR9.

Using the resize-sharpen alone will on my system use 90-95% of the CPU, but when I turn on denoise3d it goes down to 70-75% (other people have have had similar experience). This is when playing NTSC DVD from HD. When playing from external DVD-rom through firewire the CPU usage is approx. 95%.

I am using WinDVD video decoder, maybe it's less CPU hungry than yours, don't know. For lower CPU usage use bicubic resize method, and overlay if you are using VMR9.

I used intel P4 3.0C with WinXP, ffdshow(SSE2), WinDVD4.5, powerstrip (for 1360x1024 Dlia resolution), denoise 3D, resize 1360x1024, and got notice-able stutter on DVD playback.

Switched back to Win2000, same config., problem fixed.

It seems that one needs more horsepower for WinXP( We In Need of Xtra Power?).

Maybe it just my system.

People are often asking what system they should by for FFDShow use.
Well I have to say “you get what you pay for”.
The two fastest and very equal performing options are an overclocked P4 (3.5G+) on a good quality 865 or 875 Intel based main board with dual channel RAM running 1:1 timing or an overclocked Athon 64 FX with big cache and dual channel RAM.

Since the P4 solution is cheaper then an Athlon FX system, it is the natural choice for FFDShow IMHO.

The normal cheap Athlon64, with small cache and single channel RAM is NOT in the same class as the above options.

I use a cheap P4 2.8C @ 3.5G (standard Vcore and Intel cooler on only 7Volts) on a 1Gig FSB to run FFDShow with Denoise3d followed by resize (Lanczos or Bicubic, no sharpening) to 1920x1728 with VMR9 output to an overclocked Radeon 9600Pro with no tearing or dropped frames.

VMR9 softens the image, but very high resize setting removes this problem.
Double DVD resize (1440x969 NTSC or 1440x1152 PAL) with VMR9 still looks a little too soft on my display.
Obviously, running VMR9 at such high res. requires serious power.

If someone can get even close to these settings with a normal Athon 64, we would all like to here from them.

I recon if the job is worth doing, it’s worth doing properly.
You can never have to much power for FFDShow.

Regards,

Owen

___________________________
The FFDShow resize-sharpen dude.

You got lucky Owen. I can't push my 2.8c beyond 235fsb without ridiculously high vcore. It's unfortunate :(. I just ordered what are supposed to be a good batch of 3.0cs. We'll see what I get from it.

Originally posted by Owen
VMR9 softens the image

Are you sure that it's not overlay sharpening the image?
The following quote is from vpopovic's AA/AF thread.

"If I remember correctly, it was in the context of the VMR picture looking "soft" compared to overlay. His comments were centered around overlay actually had hardware based edge enhancement on most cards (to make the picture look like it had more detail to the untrained graphics card reviewer), while VMR didn't."

There might be something in this, I will see if I can find out more.

As a simple experiment, test VMR9 WITHOUT resizing then test WITH substantial resizing but WITHOUT sharpening. (Bicibic is a low ringing, none sharpening algorithum)
It is obvious that without resizing, VMR9 is VERY soft.
The higher the resize the sharper it gets.
At very high resize settings, it looks very close to overlay for sharpness but with better color.

Regards,

Owen

___________________________
The FFDShow resize-sharpen dude.

Originally posted by Owen
People are often asking what system they should by for FFDShow use.
Well I have to say “you get what you pay for”.
The two fastest and very equal performing options are an overclocked P4 (3.5G+) on a good quality 865 or 875 Intel based main board with dual channel RAM running 1:1 timing or an overclocked Athon 64 FX with big cache and dual channel RAM.

Since the P4 solution is cheaper then an Athlon FX system, it is the natural choice for FFDShow IMHO.

The normal cheap Athlon64, with small cache and single channel RAM is NOT in the same class as the above options.

I use a cheap P4 2.8C @ 3.5G (standard Vcore and Intel cooler on only 7Volts) on a 1Gig FSB to run FFDShow with Denoise3d followed by resize (Lanczos or Bicubic, no sharpening) to 1920x1728 with VMR9 output to an overclocked Radeon 9600Pro with no tearing or dropped frames.

If someone can get even close to these settings with a normal Athon 64, we would all like to here from them.

I recon if the job is worth doing, it’s worth doing properly.
You can never have to much power for FFDShow.




Dooh... I just bought a A64 3200+ system thinking that the only difference was the dual channel RAM. Didn't think it made that much difference...

Originally posted by Owen
As a simple experiment, test VMR9 WITHOUT resizing then test WITH substantial resizing but WITHOUT sharpening. (Bicibic is a low ringing, none sharpening algorithum)
It is obvious that without resizing, VMR9 is VERY soft.
The higher the resize the sharper it gets.
At very high resize settings, it looks very close to overlay for sharpness but with better color.


Stupid question but I don't remember it being asked in this context recently, but you do have zoom player fixing the VMR scaling bug, right? It defaults to off and you really need to turn it on if using VMR. At least until 9.0c comes out which fixes the bug. Otherwise the images actually get scaled down first and then up and that will make the image look soft as well. AFAIK, Zoom is the only player with a work around in place.

-Pat

Owen,

I will try that, and I don't doubt you are right about this.
However, what I am trying to say is: it seems like overlay is sharper than VMR because overlay sharpens the picture, not because VMR softens it.

Also, vpopovic discovered the same as Mark at the Immersive forums reported one year ago: you can add sharpness to VMR by using AA/AF.
I don't know if this adds real detail or not, but vpopovic's screenshots (from LOTR) showed interesting results.

Owen: thanks for the info. However, I just hate to buy intel now, knowing that all kinds of new cpu with higher pin sets, new mobo witn pci-x, ddr2, btx, etc.,etc. or coming out into the retail market. Intel is trying to put all the existing intel stuff to pasture. I guess, as a business, its good to get consumers buying, but I would like to buy something that would at least stay current for 2 or more years. What happens if and when Andy is successful to port denoise3d to 64 bits, and if it runs fast, I would kick myself for going with P4 2.8c. If this guy (no name on page 87 from Oslo, Norway, I wish he had at least given himself a code name) can do a decent job with his barton 2800+, I would expect that the regular AMD64 3200+ will do a similar job. The AMD 64 FX 53 has 1 MB cache, so does the cheap AMD 64 3200+ (it is not really cheap). My objective is to get a decent job done, and I am not a video phile that I must have the best picture possible, and in any case my sony plasma can only do 1024x768 resolution. I do not mean to be disrespectful, and I know you are the real guru here, but I just need some advice here on some middle of the road system that can last for a couple of years.

I think that's the point e... there is no middle of the road system that's going to last a couple of years if you want to keep pushing the envelope. For a stable set of parameters, sure. But If you want to be running the same settings 2 years from now, then I marvel at your self control ;)