Thanks, option C is my instinctual choice. I'd been advised by others that option A is better, just for the reason of saving settings. I would rather be safe than sorry, I hate spending hours/days troubleshooting something a proper install would have avoided.

Not to get to far off topic, but....
Owen, do you use the ffwd/rrwd options in zoomplayer? We have very similar setups (except for display devices). I lose audio sync (video ahead of audio) when fastforwarding/rewinding DVD in ZP. All other transport functions do not affect audio. I'm using intervideo codecs ver 5.2.20.37, and the built in MB audio "soundmax digital" for audio renderer.

Back on topic,

I'm going to try the latest ffdshow build tonight, see if it affects CPU utilization at all. I'll make sure to set the "Raw Video" setting :)

Originally posted by Owen

Do any 2004 builds of FFDShow work with TT, and if so which.
Since FFDShow always works with Zoom, what is going on with TT.
I am interested to know what changed in FFDshow or TT to break TT support.

Not sure what exactly broke, but the issue was with all builds after 20040329.

There is a new TT patch which resolves whatever the issue was and the newer versions are working now according to the thread over on the TT forums.

Thanks jvincent,

Its good to see that the TT developers are on top of the problem.

Owen
___________________________
The FFDShow resize-sharpen dude.

Are there any significat changes to ffdshow (as far as from a HTPC standpoint goes) from 0329 to the most current release? I like to be up to date but have noticed that ffdshow is something one must be careful with. Thanks.

Going from my bad memory, I think the "green tinge" problem was fixed and there were some performance improvements. There's probably some other stuff as well.

My HDTV is just back from a 3 week visit to the repair shop so I haven't personally tried any of the newer builds.

Originally posted by debennett2
Are there any significat changes to ffdshow (as far as from a HTPC standpoint goes) from 0329 to the most current release?

I do believe the "Process Right Half Only" option is new since 329. (correct me if I'm wrong) It allows you to apply filters to only the right half of the image, which makes for great comparisons. There are a couple new preset auto-load conditions as well.

what are the preset autoload conditions anyhow? It would be nice if I could load presets externally via a bat file or executable...individually speaking. I know it is possible to register and unregister ffdshow through a batch fine but that only works if you only have one preset you want to use, plus it could cause issues down the road if one forgot.

hello all... dusted off my HTpc recently and started updating the critical software elements... upgraded ZP to 3.31Pro, ReClock from .99b to 1.41, and ffdShow from 20030325 to 20040514. the ffdShow change did not work out at all, and it was the last software updated. (all else was working w/ the older build of ffdShow, both ZP and Reclock)...

the new versions of ffdShow seem to be causing havoc on my system, no matter what settings are applied... the output is always jerky w/ many dropped frames and seeminly playing at framerates ranging from 8-15fps (if reClocks monitor is to be believed!). i dropped back to an old version of ffdShow for now so we can watch "Hero" tonight, but i am really curious what i am missing now from the new versions featureset... is it just me, or are others having problems w/ these more modern builds of ffdShow? i tried using the same settings that have served me so well w/ the older build, so i'm not sure why the performance has degraded so... i have a 2.8c P4 w/ 1Gb of memory so my modest settings shouldn't be bringing my system to its knees like this.

is there some interraction between the newer versions of ffdShow and reClock perhaps? would love to try out this new "denoise3d" function as it sounds very helpful, but as it isn't available except in the new builds i cannot see the magicke..!

What videofilter do you use? And audio filter? I dropped Reclock when I went to the freeware AC3 Filter, it works perfectly with my WinDVD video filter. Perhaps you can drop it too? Did you try without Reclock?

Originally posted by Mastiff
What videofilter do you use? And audio filter? I dropped Reclock when I went to the freeware AC3 Filter, it works perfectly with my WinDVD video filter. Perhaps you can drop it too? Did you try without Reclock?

hello mastiff, thanks to you for your reply! i am using the WinDVD5 filter for video filter, and Sonic for audio filter (only one that has ever worked for my SPDI/F output from Audiophile 2496 properly)... i did not try dropping ReClock, as it is the only thing that has ever kept audio in-sync on my system due to my projector being locked at 60Hz. i haven't heard about this new AC3 filter though, does it work w/ DTS as well..?

the newer versions of ffdShow are very confusing to me out of the box compared to the older models, especially due to the three different control panels! is "denoise3d" worth the headache of upgrading..?

Originally posted by shinoSetsuna
the newer versions of ffdShow are very confusing to me out of the box compared to the older models, especially due to the three different control panels! is "denoise3d" worth the headache of upgrading..?

If you have the CPU power to run it, Denoise3d is definitely worth upgrading for.
Even if I did not use and other filters, I would use FFDShow just to have Denoise3d.

I recommend you try a setting of L 0.5, C 0.5, T 5.0, HQ for DVD playback.

If you go higher with L & C settings you will start to loose detail and if you go lower with T you will reduce the noise reduction performance.
Higher settings for L and C are good for reducing compression artifacts with Divx and Xvid video.

What else about the newer builds is confusing?
They are 90% the same as the old.

I have not found the newer builds of FFDShow to be slow.

Try forcing YUV2 or YV12 output color space and see which is faster on your system.

Regards,

Owen
___________________________
The FFDShow resize-sharpen dude.

Originally posted by shinoSetsuna
and ffdShow from 20030325 to 20040514. the ffdShow change did not work out at all, and it was the last software updated. (all else was working w/ the older build of ffdShow, both ZP and Reclock)...

the new versions of ffdShow seem to be causing havoc on my system, no matter what settings are applied... the output is always jerky w/ many dropped frames and seeminly playing at framerates ranging from 8-15fps (if reClocks monitor is to be believed!).

Yep, that latest ffdshow version is weird.
I tried both 01 and 08 May version, and both works well.
I tried 14 May version, it cause stutters and jerking.
Now I use 20040329 as reported by debennett2 is the ffdshow version that consume less CPU.
Yep, ready to watch DVD again.:D

after a bit of experimenting last night, i did finally find the cause of the stuttering on my system in relation to the recent ffdShow builds... it is the "resize" function, if it is enabled at all, even w/ no other settings turned on, the video will stutter and is unviewable... i tried both 1440x960 and 1024x768, but either will bring on the effect. if i leave resize off, i can use any of the other settings w/ no problem at all (and yes, i did find Denoise3d to be a remarkable function!). this confuses me because i can use either resolution for resize on the older builds w/out any stuttering whatsoever on the older builds... i am forcing the use of YUV2 colorspace, and all settings are apparently identical between each. one thing i did discover is that the newer build (20040514) does not work ever w/ the Sonic Cinemaster video codec, no matter the setting, but switching video codec to WinDVD5 will allow ffdShow to work, but no resize... confused, thinking of wiping the system clean and starting fresh tonight once it is dark enough to run the projector..!

Originally posted by shinoSetsuna
after a bit of experimenting last night, i did finally find the cause of the stuttering on my system in relation to the recent ffdShow builds... it is the "resize" function, if it is enabled at all, even w/ no other settings turned on, the video will stutter and is unviewable... i tried both 1440x960 and 1024x768, but either will bring on the effect. if i leave resize off, i can use any of the other settings w/ no problem at all (and yes, i did find Denoise3d to be a remarkable function!). this confuses me because i can use either resolution for resize on the older builds w/out any stuttering whatsoever on the older builds... i am forcing the use of YUV2 colorspace, and all settings are apparently identical between each. one thing i did discover is that the newer build (20040514) does not work ever w/ the Sonic Cinemaster video codec, no matter the setting, but switching video codec to WinDVD5 will allow ffdShow to work, but no resize... confused, thinking of wiping the system clean and starting fresh tonight once it is dark enough to run the projector..!

This is cause Athos could not compile the mplayer.dll with gcc (he got some linking errors) and had to use Visual Studio for the compiling wich in fact disables all hand coded assembler code... wich makes the mplayer rezise function rather useless and taking up to 100-400% more cpu power.

I send him a fix for the compiling problem since i could compile it without errors. Just use a older version or wait for a new build.

Originally posted by AndyIEG
This is cause Athos could not compile the mplayer.dll with gcc (he got some linking errors) and had to use Visual Studio for the compiling wich in fact disables all hand coded assembler code... wich makes the mplayer rezise function rather useless and taking up to 100-400% more cpu power.

I send him a fix for the compiling problem since i could compile it without errors. Just use a older version or wait for a new build.

helle AndyIEG, thx! is it only the 20040514 build that is so effected then, any previous build will work fine? i am surprised some here seem to be able to use this latest version w/out issue, maybe their systems are much more powerful than my weakling 2.8c P4..!

Originally posted by shinoSetsuna
helle AndyIEG, thx! is it only the 20040514 build that is so effected then, any previous build will work fine? i am surprised some here seem to be able to use this latest version w/out issue, maybe their systems are much more powerful than my weakling 2.8c P4..!

As far as i know only the latest version was compiled with VC so all older version's should run faster. Since there is a new build every 1-3 weeks some might refer to a "latest build" wich in fact is not the latest and that confuse other.

I had to go back to 20030523 version to get resize working without stuttering, I have a P4 2.66 and Sonic decoders.

Originally posted by JDLIVE
I had to go back to 20030523 version to get resize working without stuttering, I have a P4 2.66 and Sonic decoders.

hmm... after just trying the 20040508 build, i get the same result of stuttering video w/ resize function enabled that i get w/ the 20040514... from what i have seen so far no 2004 build works on my system w/ resize enabled, very strange. i wonder if there is something different in the HTpc setups such as mine and yours that is 'breaking' something in these newer builds when resize is attempted... i'm going to try a few more 2004 builds to see if the problem persists across the board on my system, and then start trying to decide which is more important to me... Denoise3d or Resize..!

Folks. FOr those of you having trouble, please do me a favor. Uninstall any current version of ffdshow you may have, reboot, install 20040329 build, enable raw video (all supported) only...don't mess with anything else. Reboot again. Now go setup your denoise and resize in that order. Is it still stuttering?

The "Horizontal line" issue of the resizer is gone in the latest 20040514 version!!!

Originally posted by hyg
The "Horizontal line" issue of the resizer is gone in the latest 20040514 version!!!

Thats not a common "FIX" its the result of the Visual Studio build of the mplayer.dll wich is also the problem why this version runs so slow. The "Horizontal line" bug is in the asm optimized code lines wich wasnt used in the latest compile.. aka only the plain c version is used wich dont have this bug, but it runs slower since no mmx/sse code is used.

I must confess that it’s hard to remember what build I am using at any point in time.
I checked my version and it is 18 April 2004 or 20040418.
I have not used any newer version but this version works great.
I’m doing a ridiculously high resize of 1920x1152 (Lanczos) with Denoise3d before resize with no problems what so ever. So this version can’t be to slow.

Owen
___________________________
The FFDShow resize-sharpen dude.

Originally posted by Owen
I must confess that it’s hard to remember what build I am using at any point in time.
I checked my version and it is 18 April 2004 or 20040418.
I have not used any newer version but this version works great.
I’m doing a ridiculously high resize of 1920x1152 (Lanczos) with Denoise3d before resize with no problems what so ever. So this version can’t be to slow.

Owen
___________________________
The FFDShow resize-sharpen dude.

only the 20040514 version has this resize speed problem, but all other versions have this horizontal line problem if u dont resize to a direct multipler of the org. res using lanczos or spline or any non linear chroma resize method.

Thanks Andy,

I was aware of that problem although I never really noticed the horizontal lines. I have found that resizing to 1152 vertically (double 576 PAL) does look better then the 1000 that I was using before.
I’m using 1920 horizontal resize and it works very well. I like it better then 1440.
Do you know if it is required to resize to an exact multiple of the source both vertically and horizontally to avoid this bug?
It looks like vertical resize is the critical one.

Has there been any progress in fixing the horizontal line problem and the “green bug” ?

Regards,

Owen

I've had the same stuttering problem you report. The problem is not present in the 20030325.

Have I the problem 'cause i resize to 960x540 where 540 is not a multiple of 16?

thanks

This horizontal line problem you mention, is that mostly evident on very bright scenes? I have horisontal lines on scenes like where Frodo wakes up in Rivendell in FotR and where Gandalf comes back to life in TT (Two Towers, not TheaterTek!).

I still don;t understand how I am able to resoze to 1776x1000 and sharp (both) at about 1.5 with a P4 2.4Ghz system (533 FSB). I have to repeat that there has to be something special about the 20040329 build over the rest. Anyone have ideas as to what that might be?

Originally posted by Owen
Thanks Andy,

I was aware of that problem although I never really noticed the horizontal lines. I have found that resizing to 1152 vertically (double 576 PAL) does look better then the 1000 that I was using before.
I’m using 1920 horizontal resize and it works very well. I like it better then 1440.
Do you know if it is required to resize to an exact multiple of the source both vertically and horizontally to avoid this bug?
It looks like vertical resize is the critical one.

Has there been any progress in fixing the horizontal line problem and the “green bug” ?

Regards,

Owen

Im still not sure seems the vertical value is the critical. Latly i just testet the 2.55 avisynth implementation in ffdshow and with lanczos4resize u get a realy great movie no green bug no horizontal lines. Im using it now but it takes 20-30% more cpu power compared to the mplayer (ffdshow) lanczos method.

I was under the impression that the green bug is gone since 09/2003 and the only green bug wich is still in is in the VMR9?

Originally posted by Mastiff
This horizontal line problem you mention, is that mostly evident on very bright scenes? I have horisontal lines on scenes like where Frodo wakes up in Rivendell in FotR and where Gandalf comes back to life in TT (Two Towers, not TheaterTek!).

Its visible in flat mono colored areas, walls doors. If u see those lines just switch to "bicublin" resize mode and check if those lines are gone. If they are gone u suffer from that resize bug and need to calc at least your vertical resize to a 1.5*org or 2*org. resolution or simply keep using "bicublin" mode. U can also use the avisynth resizers wich are slower but bugfree.

If you turn on OSD in ffdshow what does it show? I can't get anywhere near this with an overclocked P4 running 3GHz. Can't even run more than bicubic either.

Originally posted by debennett2
I still don;t understand how I am able to resoze to 1776x1000 and sharp (both) at about 1.5 with a P4 2.4Ghz system (533 FSB). I have to repeat that there has to be something special about the 20040329 build over the rest. Anyone have ideas as to what that might be?

Originally posted by AndyIEG
Im still not sure seems the vertical value is the critical. Latly i just testet the 2.55 avisynth implementation in ffdshow and with lanczos4resize u get a realy great movie no green bug no horizontal lines. Im using it now but it takes 20-30% more cpu power compared to the mplayer (ffdshow) lanczos method.

I was under the impression that the green bug is gone since 09/2003 and the only green bug wich is still in is in the VMR9?

Andy,

I think you are right. I could not notice the green but in overlay with March 29th built. In VMR9 it still exists to certain extent (looks like the higher the resize resolution, less noticable it is).

As far as Avisynth resize, are you using only Avisynth or Avisynth first and FFDShow resize latter? I am not at my HTPC to check, but if I remember correctly I had issues with using only Avisynth so that only a portion of upsized image was displayed (kind of zoom mode). I am using Forceware Multimedia Beta and March 29 built of FFDShow with Avisynt 2.55.

Horizontal line problem is pretty bad, but gets masked with VMR9 and high resolutions. I use Avisynt Lanczos to 960x640 and then FFDShow Bicubic (-0.98) 1920x1080 and then let my card scale down to 1280x720 output, VMR9. Overlay realy shows this bug.

Originally posted by vpopovic
Andy,

I think you are right. I could not notice the green but in overlay with March 29th built. In VMR9 it still exists to certain extent (looks like the higher the resize resolution, less noticable it is).

As far as Avisynth resize, are you using only Avisynth or Avisynth first and FFDShow resize latter? I am not at my HTPC to check, but if I remember correctly I had issues with using only Avisynth so that only a portion of upsized image was displayed (kind of zoom mode). I am using Forceware Multimedia Beta and March 29 built of FFDShow with Avisynt 2.55.

Horizontal line problem is pretty bad, but gets masked with VMR9 and high resolutions. I use Avisynt Lanczos to 960x640 and then FFDShow Bicubic (-0.98) 1920x1080 and then let my card scale down to 1280x720 output, VMR9. Overlay realy shows this bug.

I just use zoomplayer + overlay and resize to 1152,Y i let Y calc based on the org. resolution to get a correct AR based on 1152x.(1152 is my native monitor resolution)
I just use the latest 2.55 avisynth and lanczos4resize and all work great. The resize is realy good and clear, but i have to disable asharp or noise sometimes since i get speed problems.

Please don't flame me for this!

I built my HTPC back in January and have been happily using TheaterTek and ffdshow (Sept 2003 build) ever since. I thought I'd check in here to see what's new - looks like a lot! I'm currently using resize/sharpen (lanczos) and gradual denoise filters.

Based upon what I've read here (again), sounds like I'm due for an ffdshow upgrade so that I can use the denoise3D in place of gradual denoise. Also, I'd love to get the 1/2 picture feature to make optimization easier.

What build will give me this without a huge increase in CPU demand (my P4 3GHz is close to it's limit now!)? I am assuming the 20040418 build is what I want?

Can someone please post the link to the newer ffdshow builds. I've spent 30 minutes searching this thread and can't find a link.

Thanks,
Phil

Pmd918,

You can get the newer builds here. (http://athos.leffe.dnsalias.com/)

Completely LOST! Someone please HELP! Using zp pro v.3.31, I would like to set up and work with ffdshow. Currently I have the 20040329 version installed. When I open zp without ffdshow enabled everything seems ok. When enabling ffdshow, zp crashes and the following application error occurs:

"The instruction at "0x1c41e176" referenced memory at "0x0b9c9000". The memory could not be "written"".

The current setup uses Cyberlink PowerDVD video decoder, Intervideo WinDVD audio decoder and Default DirectSound device as audio renderer. In the ffdshow configuration ONLY raw video is enabled (under codecs). Has this error something to do with the registry? Is it a incompatibility between zp and ffdshow (or other s/w) versions? Please help.


My H/W-S/W is:
P4 3.0 GHz
ATI Radeon 9200
ASUS P4 C800 E Deluxe M/B
Infocus SP 7200 pj
WIN XP pro (sp1-updated)
WinDVD 5.3 Platinum
PowerDVD 5.0

Completely LOST! Someone please HELP! Using zp pro v.3.31, I would like to set up and work with ffdshow. Currently I have the 20040329 version installed. When I open zp without ffdshow enabled everything seems ok. When enabling ffdshow, zp crashes and the following application error occurs:

"The instruction at "0x1c41e176" referenced memory at "0x0b9c9000". The memory could not be "written"".

The current setup uses Cyberlink PowerDVD video decoder, Intervideo WinDVD audio decoder and Default DirectSound device as audio renderer. In the ffdshow configuration ONLY raw video is enabled (under codecs). Has this error something to do with the registry? Is it a incompatibility between zp and ffdshow (or other s/w) versions? Please help.


My H/W-S/W is:
P4 3.0 GHz
ATI Radeon 9200
ASUS P4 C800 E Deluxe M/B
Infocus SP 7200 pj
WIN XP pro (sp1-updated)
WinDVD 5.3 Platinum
PowerDVD 5.0

I have found the raw environment option needs to be set to RGB32 (i think) for Cyberlink codecs to work. else an error is give. Elecard and Intervideo work with it set to "all supported", and intervideo will only work i this way.

I have version 3.31 ZP, will check the FFDshow version tonight, its near the newest.

Do Greece run PAL or NTSC?

Thanks for your reply. I made the settings you said, but unfortunately the problem exists! I think there should be a setting in registry, but I don't know what... Greece works on PAL. Thank you in advance

Yes for PAL there needs to be for Intervideo

Its under intervideo - common - video and maybe videodec (i added to both)

DXVAPAL with the dword set to 1

possibly a stupid question, but...

Is there a way to pre-process a movie with ffdshow so that the maximum massaging can be accomplished with even a less powerful processor that could not accomplish the task in real-time?

As hard drive space is so cheap now, I certainly wouldn't mind starting a movie processing in the morning for an evening view.

That would be a really cool thing if posssible! I'd love to see what I can do given enough time to process the video!

I guess this would involve processing and re encoding of the video? As FFDShow is inserted after the codec and is dealing with the raw/decoded video.

Re encoding would induce some loss (i guess high birates could be used to reduce/stop this). On the other hand massive amounts of processing can be given to the task, say a 2hr movie could be processed with FFDShow for 6 hours, (one would need to also allow for re encoding).

Is it really worth it but? What would the gains be?
Can any current PC FFDShow (with worthy settings) HD matterial, surely even HD can benefit, i doubt its possible on the fly at present, but pre processing could work?

would jsut be intersting to find out at least...

Based upon what I've read here (again), sounds like I'm due for an ffdshow upgrade so that I can use the denoise3D in place of gradual denoise.

Good luck, just turning on that one feature sends my computer to fits with 100% cpu useage. Even without resizing. On the other hand, I can resize all the way to 1440x960 and beyond with lanczos and stay at around 70% with my 2.5gig p4 if I dont have it on.

The biggest upgrade for me was getting sonic decoder and resizing to double dvd resolution 1440x960 with denoise at about 30. I dont need to do anything else this is as good as it gets. Almost as good as a movie on HBOHD.

Before, I had windvd and powerdvd and resizing at 1280X720, my native resolution but the picture looked soft and colors were cartoonish. A very aweful picture in comparison.

Brian

Originally posted by lescaudle
possibly a stupid question, but...

Is there a way to pre-process a movie with ffdshow so that the maximum massaging can be accomplished with even a less powerful processor that could not accomplish the task in real-time?

As hard drive space is so cheap now, I certainly wouldn't mind starting a movie processing in the morning for an evening view.

Sorry, there is no practical way.
To proccess the video and resize to something like double DVD resolution and save it to disk uncompressed would need about 180Gig per hour.

Regards,

Owen

___________________________
The FFDShow resize-sharpen dude.

Owen - in my fantasy world, I could process using ffdshow to create an HDTV transport stream file that my MyHD card could play in 1080i. Only takes 8GB per hour.

But even at 180 GB per hour, that becomes quite possible within next year or 2 as hard drives swell to 400 and 800 GB.

Of course, by then, 'they' will have weaned us off DVD onto HDTV blue-ray whatevers with lots of encryption - and we'll learn to enjoy 30 minutes of forced commercials before viewing a movie once - before it self-destructs.

Well, maybe not!

Last night was my first time using FFDShow. Started with "Post processing with FFDShow by Joshus Schipmann" (great step by step setup guide) and PQ look like crap (almost the opposite of his before and after shots). I then set to see effect on right side of screen. Played with lots of adjustments but the left side (original) always look superior to the processed side. Also when I would set resize to 1360x1024 the pic would freeze on the first frame so I set "expand to next multiple of 16".

Does anyone have some settings suggestions for my setup?

DILA (G11) w/ISCOII at 1360x1024
3 GHz, 512MB, Radon 9200

William,
Try resizing with Lanczos to either 1440X960 or better yet 1920x1080. Your computer should handle that nicely. Play with the luma, etc sharpening a bit. Use gradual denoise about 30-40 to taste and try denoise3d instead and see if your computer is stable with that. Also use vmr9 verus overlay and see if you are stable with that. You did not mention what software decoder you were using. I still prefer sonic dvd.

Brian

Originally posted by usabrian
William,
Try resizing with Lanczos to either 1440X960 or better yet 1920x1080. Your computer should handle that nicely. Play with the luma, etc sharpening a bit. Use gradual denoise about 30-40 to taste and try denoise3d instead and see if your computer is stable with that. Also use vmr9 verus overlay and see if you are stable with that. You did not mention what software decoder you were using. I still prefer sonic dvd.

Brian
Brian,
I'm using ZP with WinDVD 5 filters. So it is OK to resize larger than your display? I will try in a little while.

Yes. Your graphic card will scale down the picture.

Anyone tried the 0520 build?

I currently have an AMD 2500 Barton. What do I need(cpu) to resize to 1776x1000 which is my desktop res?

simply Lanczos resizing to 1776x1000 is already pretty CPU intensive. I'd say a P4 3+GHz would be needed.

I just got FFDShow. Before I install it I wonder if I should uninstall all codecs in my computer before.

Thanks,
Santiago

:confused: :confused:

Filter Connection Error

(Sonic Cinemaster DS Video Decoder). Video Out ---->(FFDShow Video Processor). In

Not All Filters Can connect properly

I have ffdshow 2004 configured as always. I am using the same zp version as always but suddenly I get this message. If I turn off ffdshow in zp it plays fine. I have all codecs disabled in ffdshow except raw.
What am I doing wrong anyone?

create a DWORD registry value under [HKEY_LOCAL_MACHINE\SOFTWARE\Sonic\CineMaster DS DVD\2.5\VideoDecoder] called "AllowAllRenderers" and set it to 1

Although I 've gone through all 53 pages of this "ffdshow faq" I haven't found anyone having the problem I face right now: I installed zp and currently use WinDVD video decoder, default audio renderer (soundmax built in ASUS P4C800E Deluxe m/b), ffdshow and as audio decoder I tried AC3 filter as well as WinDVD audio filter. I want to use the SPDIF output to have my A/V receiver do the decoding. Both audio decoders seem to work fine, until the moment you press pause or rewind/forward the movie. From that moment there is no sound and my A/V receiver turns into PLII mode. To get it right again all you have to do is restart zp. This doesn't happen when you change chapters. Has anyone encountered such a problem? Thanx again guys for the info you provide in here, without you I wouldn't even know that these tools exist!!

Sorry for giving you incorrect info. Unfortunately same thing happens even when changing chapters. Also, same problem occurs even when I use reclock (default settings) as audio renderer. Any idea, please?

Good Day!

Can you help me to find some manual where is explanation about how to watch external subtitles with DVD movies?
In case movie.avi there is no problem - just rename subtitle.srt in movie.srt

but..what about DVD?

Please suggest the best FFDShow settings for me to use with the following:
Zoomplayer+Sonic1.5

Is VMR or Overlay better with Sonic?

Is Lanczos, Spline, or Bicubic better with Sonic?

Would using DVDIdle allow me to theoretically increase the settings in FFDShow?

Originally posted by JoeFigueiredo
Please suggest the best FFDShow settings for me to use with the following:
Zoomplayer+Sonic1.5

Is VMR or Overlay better with Sonic?

Is Lanczos, Spline, or Bicubic better with Sonic?

Would using DVDIdle allow me to theoretically increase the settings in FFDShow?

Q: Is VMR or Overlay better with Sonic?
A: depends on your settings and videocard (just try it and see what looks better for u)

Q: Is Lanczos, Spline, or Bicubic better with Sonic?
A: Lanczos is the most used and best tradeoff beetwen cpu usage and scaled image quality (it also dont matter what decoder u use)

Q: Would using DVDIdle allow me to theoretically increase the settings in FFDShow?
A: no

I have an NVidia 5900XT. I can't decide what's better between VMR and Overlay (VMR looks smoother, but Overlay has more vibrant colours (however I can't tell if this is because my color settings need changing between switching them.)

Another issue I have:
I see lots of people in here saying they can get up to 70% CPU when using FFDShow. I've never gotten my CPU that high (according to Task Manager). My FFDShow settings will freeze the picture when set to high, but my CPU will still only be at about 55% max peak. I'll get studdering using Sonic decoders when FFDShow is set high, but again CPU is at 55% max. Is this normal?

I can run pretty much on 100% CPU with my XP2400 and still remain smooth. (WinDVD or Elecard codecs)

I had 70% CPU limitation on tearing, but recently have reinstalled my player (FM 3.0 Beta), FFDShow, Avisynth, video and audio drivers and DX 9.0b and switched to single monitor configuration. Now I get with 95% CPU utilization in VMR9.

Having 55% CPU limit could be normal under certain circumstances. There are two main limiting factors on FFDShow performance. CPU and RAM speed. You might be OK on the CPU, but you might be using single chanel slow RAM, so your CPU will have to wait for your RAM to catch up and waiste its cycles. Your RAM timings and settings in BIOS might be either wrong or very conservative (for stability reasons). If this is not the case, try de-installing all components listed above and re-install them.

With Nvidia cards you are probably better off using VMR9. I don't think Nvidia uses (or at least I can't see the full benefit of it) 128-bit floating point color rendering that your card is capable of. Color rendering performance also depends on how your system builds the graph for playback, which believe me is a mistery.

With VMR9 and single monitor setup (be sure you boot in that setup just in case, drivers and DX9 sometimes don't change the graphs) your colors should be, well, cinematic. Although they look better saturated in overlay, and you have the convenient color controls, VMR9 in my setup renders them better (i.e. using higher precison calculation for more shades of the same color which gives the image colors close to cinema quality). You can compensate saturation either by adjusting it on your display, or in FFDShow. You can not compensate rendering. It is done only once and what has been done is done. Softness in VMR9 can be compensated with Lanczos scaling in FFDShow. You should try all algorithms and see which one you like.

I was getting the green garbled mess like about 6 others in this thread. I got it using zp, ffdshow, nvidia (nvdvd) for audio and video. I got it w/ overlay and vmr9, as well as every other option I could think of. What finally fixed it, for me, was to install vobsub. Someone else in this thread already suggested it.

I didn't configure vobsub at all.

If i have vobsub below ffdshow in the zp filter list, i get a green garbled mess along with a huge green line across the bottom. If i put ffdshow on bottom, then all seems to work well. After it was working, I tried setting resize to 1024x768. It looked good, but was slightly choppy.

I didn't feel like trying to find a resolution between native and 10x7, so i shut it all off and started NVDVD in hardware mode w/ overlay to compare and see if i'd made any progress. I think nvdvd in hardware mode looked better than anything else, but maybe its just too late for tired eyes.

Vmr9, in both software and hardware, w/ zp and w/ nvdvd, seems too warm, and while the colors are VERY saturated, there is no sharpness at all.

Originally posted by vpopovic
Having 55% CPU limit could be normal under certain circumstances. There are two main limiting factors on FFDShow performance. CPU and RAM speed. You might be OK on the CPU, but you might be using single chanel slow RAM, so your CPU will have to wait for your RAM to catch up and waiste its cycles. Your RAM timings and settings in BIOS might be either wrong or very conservative (for stability reasons). If this is not the case, try de-installing all components listed above and re-install them.

With Nvidia cards you are probably better off using VMR9. I don't think Nvidia uses (or at least I can't see the full benefit of it) 128-bit floating point color rendering that your card is capable of. Color rendering performance also depends on how your system builds the graph for playback, which believe me is a mistery.


I have 512 DDR PC3200 400MHz RAM and it is set properly in the BIOS (maybe someone can clarify with this RAM what my BIOS should be set to?).
I've reinstalled numerous times all apps but with the same results.

To reiterate and clarify: For my PC config, I can set fairly high settings in FFDShow (Lanczos 1440x960, Denoise3d, DScaler Sharpening) and play them fine with minor studdering, but my CPU utilization in Task Manager varies from 45-55%. What I'm wondering is why it's not up around 90%, as again i am having studdering problems that high. Theoretically, if my CPu is only at 55%, I should be able to push FFDShow even higher.

Originally posted by JoeFigueiredo
I have 512 DDR PC3200 400MHz RAM and it is set properly in the BIOS (maybe someone can clarify with this RAM what my BIOS should be set to?).
I've reinstalled numerous times all apps but with the same results.

To reiterate and clarify: For my PC config, I can set fairly high settings in FFDShow (Lanczos 1440x960, Denoise3d, DScaler Sharpening) and play them fine with minor studdering, but my CPU utilization in Task Manager varies from 45-55%. What I'm wondering is why it's not up around 90%, as again i am having studdering problems that high. Theoretically, if my CPu is only at 55%, I should be able to push FFDShow even higher.

I'm sure it's somewhere earlier in the thread but task manager basically averages your CPU utilization over the sampling period (1 second). It also combines multiple "CPU's" into a 100% total. If you have hyperthreading on, there's a good chance you're really running one thread of execution at close to 100% (the entire video processing chain happens in a single thread). It's also possible (though probably less likely) that you are spiking at higher utilizations and the dropping frames really brings it back down so the "average" you see is artificially reduced by the stuttering itself.

Is there a utility I can use to provide more accurate CPU utilization with a P4 HT processor?

I suspect that ffdshow either isn't multi-threaded to take advantage of HT technology, or that it's not possible to do so due to limited resources within a HT-enabled CPU. That's what is keeping the CPU utilization pegged at 50% on only 1 virtual processor. I'm not sure of this though, it's just my theory.

Regarding the task manager sampling rate, that's another problem altogether. I used to trust the CPU utilization shown by the ffdshow OSD a lot more since the sampling rate seems a lot higher, but I've heard of forum members advising against trusting those figures as well.

I bet if you turn off Denoise3d you will see all that choppiness go right away. I love what it does to the picture but its bite of processor power is just unacceptable.

Brian

I tried the latest alpha (ffdshow-20030523.exe) and after turning it down, turning it down, I am only left with resize at 1440x960 enabled and bicubic or experimental gives same result, still freezes. I am running a 2200+ XP with 512 dual at 400Mhz. Windvd 5 video decoder.

Can anyone recommend possibly another build that I might see better results with? I have turned off all processes I can, my utilization is approx. 93% before it freezes.

Originally posted by bedo
I tried the latest alpha (ffdshow-20030523.exe) and after turning it down, turning it down, I am only left with resize at 1440x960 enabled and bicubic or experimental gives same result, still freezes. I am running a 2200+ XP with 512 dual at 400Mhz. Windvd 5 video decoder.

Can anyone recommend possibly another build that I might see better results with? I have turned off all processes I can, my utilization is approx. 93% before it freezes.

I would not be surprised at all if 1440x960 wasn't just way too much for a 2200+ XP to handle. ffdshow generally performs better on P4's and even there you need to be in the 3Ghz range to do much more than just resize and simple noise reduction.

Originally posted by Goi
[B]I suspect that ffdshow either isn't multi-threaded to take advantage of HT technology, or that it's not possible to do so due to limited resources within a HT-enabled CPU. That's what is keeping the CPU utilization pegged at 50% on only 1 virtual processor. I'm not sure of this though, it's just my theory.


It's probably outside of ffdshow's hands for the most part. Dx handles the handoff from filter to filter and it probably wouldn't react too well getting things on different threads (I could easily be wrong here). Handling images on multiple threads is pretty complicated to do right and may potentially cost more than you get out of it in handoff latency between the threads. You could process different parts of the image in diffferent threads or do different processing steps in different threads. Given the proliferation of HT processors and the migration to multi-core processors in the future for both AMD and Intel it's probably going to become more important to look at.

I would not be surprised at all if 1440x960 wasn't just way too much for a 2200+ XP to handle. ffdshow generally performs better on P4's and even there you need to be in the 3Ghz range to do much more than just resize and simple noise reduction.

Thanks for your reply. I plan to play around tonight with different (lower) resizing.

Originally posted by pcgeek
It's probably outside of ffdshow's hands for the most part. Dx handles the handoff from filter to filter and it probably wouldn't react too well getting things on different threads (I could easily be wrong here). Handling images on multiple threads is pretty complicated to do right and may potentially cost more than you get out of it in handoff latency between the threads. You could process different parts of the image in diffferent threads or do different processing steps in different threads. Given the proliferation of HT processors and the migration to multi-core processors in the future for both AMD and Intel it's probably going to become more important to look at.

I asked Blight (ZP developer) if he could "assign" different virtual CPUs to various filters/programs in the graph and he said it is really application level issue. Based on posts in couple of threads it looks like FFDShow and Avisynth are not coded to take advantage of HT. Also, based on some posts, they don't do well on dual CPU (can't do smooth playback). My hope was that dual CPU setup would be cure to current limitations of CPUs.

lescaudle,

Re: Preprocessing video.

I'm definitely no expert in this but.... you can frame serve using AVISynth filters and do many of the functions in ffdshow. It is done by the SVCD crowd to correct and compress video but can go the other was as well. I have no idea as to what the result would be like. I managed to get ripped DVD files to frame serve to Premiere and then out to an encoder. I even was able to "scrub"! There is a lot of information on frame serving at http://www.doom9.net and http://www.videohelp.com among others. The disk space for possessing is only about twice the size of your input files since a single frame at a time is decoded - processed - encoded.

Charlie

I am using a P4 2.8C, with 512DDR PC3200, Nvidia 5900XT@1280x720 desktop resolution, using Zoomplayer with sonic 1.5 video decoder, intervideo audio decoder, reclock audio renderer, and of course, FFDShow.
This is all being fed to a Sony 42" GWIII TV via DVI.

Stretching my hardware to the limit, I can achieve smooth DVD playback (occasionally with a studder) on any one of the following setups:

Setup A:
VMR9
FFDShow in order:
1. DScaler Sharpen = 128
2. Gradual Denoise = 30
3. Resize = 1440x960 with Lanczos@10 and L&C each at 1.5

Setup B:
VMR9
FFDShow in order:
1. DScaler Sharpen = 128
2. Denoise3D = 0.5, 0.5, 5, & HQ
3. Resize = 1280x720 with Lanczos@10 and L&C each at 1.5

Setup C:
Overlay
FFDShow in order:
1. DScaler Sharpen = 128
2. Denoise3D = 0.5, 0.5, 5, & HQ
3. Resize = 1440x960 with Lanczos@10 and L&C each at 1.5

Setup D:
Overlay
FFDShow in order:
1. Resize = 1280x720 with Lanczos@10 and L&C each at 1.5
2. Gradual Denoise = 30
3. DScaler Sharpen = 128

General observations:
A. I find that 1440x960 is much better vs. 1280x720(my desktop res) when resizing in FFDShow. It is probably the one setting that has the most dramatic differences in picture quality.
B. Setup D is the most stable (no studderings in the video).


Questions:

1. Can you tell me for my hardware setup what theoretically should provide the best picture quality? I'd like to know how some of the experienced tweakers here would rank the 4 setups in order of picture quality.

2. I've been told that with my Nvidia video card that I should be using VMR9 instead of Overlay, but VMR9 takes away processing power from other options I can use in FFDShow. What is better, to use VMR9 or Overlay with higher settings in FFDShow?

3. Postprocessing option in FFDShow - what is the general consensus on this option. It seems to improve picture quality when left with only the MPlayer option, but are you losing detail with it?

4. Lastly, I know it's better to denoise and sharpen after resize, but:
a. how does that weigh against a higher resize and therefore having to denoise and sharpen before resize?
b. is it better to denoise first and then sharpen, or sharpen first and then denoise?

Sorry for the long post, but after reading every post in this thread, these are my unanswered questions (even after experimenting [can't decide]).

Originally posted by JoeFigueiredo [/B]

Questions:

1. Can you tell me for my hardware setup what theoretically should provide the best picture quality? I'd like to know how some of the experienced tweakers here would rank the 4 setups in order of picture quality.

Dude just use what look's best for your eye's, u arnt tweaking the setup for the forum u doing it for yourself. In general resize to 1440 + denoise3d will look best.

2. I've been told that with my Nvidia video card that I should be using VMR9 instead of Overlay, but VMR9 takes away processing power from other options I can use in FFDShow. What is better, to use VMR9 or Overlay with higher settings in FFDShow?

Old discussion, some prefer overlay some vmr9 just play around with booth and try set gamma/hue ... correct than choose whatever looks better for u.

3. Postprocessing option in FFDShow - what is the general consensus on this option. It seems to improve picture quality when left with only the MPlayer option, but are you losing detail with it?

Normaly u dont need those pp filters from dvd playback, they will help with crappy rips/encodes. U can try to set costum mode and only enable dering, with mplayer enabled. Dont enable nic or anything other.. no deblock since nic or mplayer deblock smooth details away. Set strength to 100-200 and just dering with mplayer algo, wich has only a very small impact but helps with ringing.


4. Lastly, I know it's better to denoise and sharpen after resize, but:
a. how does that weigh against a higher resize and therefore having to denoise and sharpen before resize?
b. is it better to denoise first and then sharpen, or sharpen first and then denoise?

resizing to 1440 should be the first goal than choose sharpen or dering. Sharpen works best after resize while dering can be also done before. My advice use denoise3d than resize to 1440 and try add a sharpen filter (i like using asharp with low settings since the resizer already sharp the picture)

to b. always denoise before sharpen

Thanks AndyIEG. I'm trying to decide for myself which has the best pq, but without a side-by-side comparison, it's very difficult, hence why I'm trying to get others' opinions.

Anyone else have some advice?

I just noticed a strange ffdshow issue on my setup. Some content was having skips. I won't say it was studdering because the audio was perfect. Th video will skip ahead a frame or two. At first, I thought it was CPU or because I recently upgraded my ffdshow to one of the May 2004 releases. Now, I'm back to the 03/29/2004 release, but I still have the same problem. I think I've traced the problem back to video vs. film sources. Film sources work just fine. Video has a problem.

My setup:

FFdshow:
1. Denoise3D = 0.5, 0.5, 5, & HQ
2. Resize = 1280x720 with Lanczos@10 and L&C each at 1.5

DVD Player: TheaterTek 64
CPU: P4 2.8Gig w/512 Meg RAM
Video: ATI 9600 (non-pro)
Audio: M-Audio AP2496

My CPU, according to XP's task manager, is only running around 40%. Based on reading this forum, I figure my CPU was plenty for what I'm doing. Is the extra frame rate of video my problem? If so, is there a better way to automatically turn things off for video content. I'm trying to keep my HTPC pretty much like a STB, but having to pull out my keyboard and mouse for video content ruins that option.

Joe,

Couple of comments:

With Nvidia overlay you might be using 8-bit color processing while VMR9 will for sure give you 128-bit floating point precision.

In my experience software resize to 16:9 format gives you marginaly better result than doing that conversion in hardware.

But again, use whatever you like the most. Also, you might like different setups for different movies.

If the hardware (the video card) is doing that resize when in VMR9, shouldn't I be able to turn up FFDShow settings higher than with Overlay, as overlay uses the CPU where VMR9 would use less cpu and more graphics cpu?

Overlay uses vid too, just in a different way. It all depends on the drivers, so I guess nVidia and ATI could vary wildly. I can't speak for nVidia, but I know ATI uses a bit more cpu on it's 95/96/97/9800 series cards for VMR9. I do, however use VMR9 because I'm convinced it operates in a wider colorspace.... on some scenes I can see banding and/or dithering artifacts using overlay that aren't present using VMR9.

In ZP, since you can't access colour controls from the shift+O manner. Can I just use the same values in the Options screens that Overlay has (that is, for Brightness, contrast, etc.) Or does VMR9 have different settings typically?

Overlay mixer (as renderer in DVD graph) is hardware supported in recent video cards, while VMR9 (as renderer in DVD graph) is software supported. In this configuration, VMR9 will always use more CPU than overlay.

FFDShow resize does not have anything to do with video renderer. It will do the same job whether you have VMR9 or overlay connected to the graph. Once FFDShow is done with postprocessing, it outputs to your video card and further processing/rendering of the signal depends whether you use overlay or VMR9.

The best results I ever got from VMR9 on any HTPC was during my recent reinstal of all audio/video related components and drivers, when I get no tearing or stuttering with 95% CPU utilization. In overlay, I can still do the same FFDShow postprocessing with 80-85% CPU utilization. With current OS and hardware VMR9 is always more CPU consuming than overlay. That will be true until video cards and OS support some kind of VMR9 hardware acceleration. From what I hear, this will happen with "VMR10" in Longhorn where "DX10" will solicit VMR9 acceleration from GPU. DX9 does not support that function.

Well, done some more experimenting, and decided to O/C my P4 2.8 10%. This allows me to get the following config. with no stuttering at all (can't tell the CPU as I have HT enabled). It is the best stable pq I have ever achieved:

VMR9 in ZP with Sonic video decoder
In FFDShow:
1. Denoise3D @ 0.5, 0.5, 5 & HQ
2. Unsharp Mask @ 20
3. Lanczos resize to 1440x960 with L&C both at 1.3

Simply amazing pq!

Joe,

We heard that your P4 could go to 3.5 Ghz. You might be able to push it a bit more.

By the way, good news for all FFDShow users is that new DVD transfers might come noise free, so you will gain some CPU power. I just saw recently released (in US) Angel Heart Special Editon and all I can say is that noise is gone. Digitally removed in a way no FFDShow could do it. One of the best DVDs in my collection. I have the first edition of the same movie, and that was noisy as hell.

Would noise-free DVD transfers mean that you can essentially shut off things like denoise3d? Or would you still use denoise3d and it's just that no noise will mean the CPU would by default not have to work as hard?

Which most recent version of FFDShow would be most optimized for a P4 2.8C HT CPU?

Originally posted by vpopovic
Overlay mixer (as renderer in DVD graph) is hardware supported in recent video cards, while VMR9 (as renderer in DVD graph) is software supported. In this configuration, VMR9 will always use more CPU than overlay.


Hmmm... are you sure?

http://msdn.microsoft.com/library/default.asp?url=/library/en-us/directx9_c/directX/htm/vmrsystemrequirements.asp

Seems to me VMR9 exposes just as much offloading ability as overlay, I just don't think the manufacturer's drivers are up to the task yet. I do want to re-iterate my perceived diferences between overlay and VMR9 rendering in terms of colorspace, however. I think there may be significant differences when it comes to ATI cards, but more research is warranted. I know for sure I see banding using overlay that isn't present using VMR9, and it makes me wonder if it's because the 95/96/97/9800 series perform all color operations in a 128bit space on the d3d path. From extremetech:


Each color channel (RGBA) gets single-precision floating point. ATI considered a 64-bit implementation where each channel would be have what's called a "short float" precision of 16 bits, but ATI instead opted to shoot higher and went ahead with a 128-bit implementation. This 128-bit precision is maintained throughout Radeon 9700's entire rendering pipeline, right up until the RAMDAC scan-out, where it's dithered down to 32-bit.


Obviously the RAMDAC doesn't matter in this case if you're outputting using DVI. I'll post some screenshots when I get the time... more to follow...

This may be the stupid question of the day, but, "Do I need the higher resolutions if I'm using a pani 301?" I'm asking because I plan to do the unthinkable, build a HT/Gaming PC. An o/c AMD64 2800 would set me up well for gaming, but the consensus is you need a o/c P4 2.8c for ffdshow. I watch TV and DVDs as much as I play games so both are equally important.

Bottom line, since my pj does not support high resolutions, should I get the AMD64 for the gaming or should I get the P4 because I'll need to run high resolutions in ffdshow and the pj will downsize? Or is something else the issue?

Thanks in advance

You don't need an o/c P4 2.8 to watch dvd, only if you want to do extensive post-processing in FFDShow (which makes DVDs' pq very good).

You could not run FFDShow and still watch DVDs, in which case all you'd need is a 1Ghz PC.

It really depends on how obsessed you are about pq (like me) ;)

A 3+GHz P4 system is an extremely capable gaming machine, so I'd say go with the P4 rather than the A64. Of course, the A64 is better for gaming, but it also depends on your graphics card, and I'm not sure of ffdshow performance on the A64.

Originally posted by JoeFigueiredo
Which most recent version of FFDShow would be most optimized for a P4 2.8C HT CPU?

Anyone able to answer this question for me?

Originally posted by JoeFigueiredo
Anyone able to answer this question for me?

"Originally posted by JoeFigueiredo
Which most recent version of FFDShow would be most optimized for a P4 2.8C HT CPU? "

I already posted a long answer regarding those "P3" or "P4" optimizing requests in the doom9 ffdshow dev thread.

The answer is there is no special version for P3 or P4 or HT wich actualy runs faster than a simple "blend" version....

The so called "P3" version simply use some few PPro instructions wich wont effect the performance, since all the big calculations are done in hand coded asm parts and arnt effected by any compiler switch.

So just use whatever version u like all will perform the same on your CPU.

Originally posted by TruckChase!
Hmmm... are you sure?

http://msdn.microsoft.com/library/default.asp?url=/library/en-us/directx9_c/directX/htm/vmrsystemrequirements.asp

Seems to me VMR9 exposes just as much offloading ability as overlay, I just don't think the manufacturer's drivers are up to the task yet. I do want to re-iterate my perceived diferences between overlay and VMR9 rendering in terms of colorspace, however. I think there may be significant differences when it comes to ATI cards, but more research is warranted. I know for sure I see banding using overlay that isn't present using VMR9, and it makes me wonder if it's because the 95/96/97/9800 series perform all color operations in a 128bit space on the d3d path.

My statement was overly simplified. You are right in saying that VMR9 offers the same hardware offloading (and even more) as overlay. In fact VMR9 can use DirectDraw (overlay) surfaces for rendering. When I said "software" I realy meant DX9. Direct draw is also a software protocol, but much simpler than DX9. In other words, VMR9 realy uses more CPU resources than overlay. Among other things CPU has to calculate the image differently for processing in 3D portion of the video card. As I understand it, instead of just taking information on per pixel basis from decoder, it will have to recalculate it and serve video as 2D polygons to GPU. Then GPU will process and render the image using 128-bit color precision and sub-pixel precision (as I understand ATI uses 4 bit sub-pixel precision, while Nvidia is quiet on their specs for FX series, but for Quadro series it is 12 bit).

Again, it looks like it is important to have "pure" VMR9 mode for color fidelity and avoid any use of overlay surfaces.

Originally posted by AndyIEG

I already posted a long answer regarding those "P3" or "P4" optimizing requests in the doom9 ffdshow dev thread.

The answer is there is no special version for P3 or P4 or HT wich actualy runs faster than a simple "blend" version....



Ok then, which build is the fastest, of late?

Originally posted by JoeFigueiredo
Ok then, which build is the fastest, of late?

... there is no fastest, just use the latest from http://athos.leffe.dnsalias.com/

The mention of video card "hardware offloading" indicateds DxVA with Overlay.

Once FFDShow is enacted i have found enabling/disabling DxVA has no effect, and hence i assumed DxVA was possible as the "link" between the codec (ie Cyberlink/intervideo) was not direct, but instead with FFDShow inbetween.

i am using zoomplayer with ffdshow and dscaler filters.....and for some reason on dvds such as pirates of the caribbean and matrix revolutions the dvd will stop playing after like 11 minutes.....but with other dvds its has no problems....and my video decoder is elecard which i believe is much better than all of the other ones........i use resize, blur, temporal comb filter(dscaler), and sharpen......please if any could help with problem it would be fantastic

i am using zoomplayer with ffdshow and dscaler filters.....and for some reason on dvds such as pirates of the caribbean and matrix revolutions the dvd will stop playing after like 11 minutes.....but with other dvds its has no problems....and my video decoder is elecard which i believe is much better than all of the other ones........i use resize, blur, temporal comb filter(dscaler), and sharpen......please if any could help with problem it would be fantastic

I just noticed that the 05-20-2004 version has the long-standing green level/resize bug fixed -- or at least extremely mitigated. In the Avia color decoder check, green is only down 5% now after doing resize instead of 25% like it used to be. Finally color balance is more or less correct with resize/sharpen.

I've spent a quite a few hours over the past week, poring through the messages on various fora. In many cases, later posts contravene many of the original suggestions. I have to admit that the pervading theme is, "Here are some worthwhile settings; start with those, and then tweak from there." I can accept that, but there are still two major items that are quite ambiguous to me. BTW, I'm using TheaterTek 1.5(64) and ffdshow 20040520.

Sharpen versus Resize

There seem to be two general camps: This thread suggests using Resize with Luma and Chroma set to 1.30, followed by Gradual Denoise (set to 30). The main thread on TT's forum suggests using Gradual Denoise (set to 30-40) followed by Sharpen (with Unsharp Mask set to 22).

Overall, the Resize seems to look better to my eyes, while the Sharpen option looks more artificial. However, even my 2.4GHz P4 w/1GB of memory doesn't have the speed to do the resize at either 1440x960 (double-DVD resolution) or 1280x1024 (monitor resolution). This is regardless of whether gradual denoise is enabled (although it's less jerky with it off).

Since I'm hoping to pass a hi-res signal to my RPTV (once I get everything else configured in my office), there's no point in doing a smaller resize, as it will need to be resized again by hardware (either in my computer or TV) before it's actually displayed on the TV. Therefore, I might have to live with the Sharpen option, even though it doesn't look as natural. I wonder, though... What type of machine would be required to do the Full Monty, if mine isn't nearly capable?

Aspect Ratio

The issue of aspect ratio is a confusing one. In ffdshow you tell it to do "No aspect ratio correction", because TT supposedly handles this. In TT's AR menu (Ctrl-E, for those of you who don't know), you click on the current mode (in blue), press [Reset], and then uncheck the "Keep Aspect Ratio" setting. Then you're supposed to set your aspect ratio, press [Save], and the whole world will be hunky dory.

This is where I'm confused. When I try this, I think the various Aspect Ratios look "strange". People are too skinny or fat, etc.

In 16:9 and Pan&Scan, unchecking the KeepAR setting causes the image to fill the horizontal dimensions of my computer monitor. For 4:3, the image doesn't reach any edge. The blue lines above and below always stay in the same position, and are nowhere near the top and bottom of my 4:3 monitor.

For 4:3, should I assume that it's already the proper ratio on screen, and just use the "Bigger Both" button to increase it to fill the screen in either dimension?

With 16x9, is it good as-is, or should I expand the height so that the blue bars just disappear off my monitor?

And is Pan&Scan supposed to be 16:9 or 4:3? I've never really understood what P&S was.

Or should I be grabbing a tape measure and physically measuring the displayed image, so that it's literally 4:3, 16:9, etc? If that’s the case, my question regarding Pan&Scan is even more pressing.

Oh yeah...

I'm also a little confused by the "Interlaced" settings here and there in ffdshow. Should I assume that DVD content is interlaced or non-interlaced?

Some people use the descaler filter with ffdshow, but I don't seem to have any filters available in the configuration settings. Where do I get these? It may be a mute point, though, considering this would require even more horsepower from my CPU.

Also, I can never seem to find a straight answer regarding resolution. I know that my Toshiba 65HDX82 widescreen RPTV will do 480i, 520p and 1080i vertically. However, what should the horizontal resolution be? 1920 seems to be mentioned often, but I've also heard the 1440 (double DVD resolution) is also worth considering. Is there as "standard" for this? Also, how does overscan factor into this?

One final question: Will ffdshow automatically be used when playing back various video material using Window's Media Player? If so, then I would like to use it on my main machine, where I'm often viewing crappy little movies from the web, which could really use some visual enhancement. Does this happen automatically, or do I need to enable ffdshow as a filter? I've hunted about a bit, but haven't found any obvious settings yet.

Thanks for any help that you guys can provide. I've also posted this on the TT forum, in case you want to take the conversation over there.

-=> Mike Hanson <=-

hi all,

i hope this wont crash my lycos account... but i need more test feedback.

I did some optimisation to the ffdshow code and need some tester's.

I mainly need ppl with P3 and P4 to test.

So if u dont worry tying a testversion here u can get it:

http://mitglied.lycos.de/ieggei2/ffdshow/

The main improvment is the optimized denoise3d filter (only the HQ version for now) so if u can test this version against your old version pls?

Btw try to use the denoise3d now AFTER the resize maybe.

Test it for bug's/anomalys pls, maybe give me feedback on how much faster the new version is compared to your old... i know its hard to test but gimme a hint.

thx

PS: pls only download it if u realy want help with testing... its a lycos account and i dont have 1GB traffic...

Could somebody please tell me if I should use a Prescott or a non Prescott P4 3.0 for overclocking to ffdshow use? I'm about to build a new system for this use, and I'm kind of confused. I have been an AMD man for years now... Are there any advantages in using the Prescott? I have heard from overlocking experts that the non Prescott is better because it can be pushed farther without getting too hot.

I'll load it up this week Andy and see if it lets me denoise after. That would be outstanding.

Originally posted by Mastiff
Could somebody please tell me if I should use a Prescott or a non Prescott P4 3.0 for overclocking to ffdshow use? I'm about to build a new system for this use, and I'm kind of confused. I have been an AMD man for years now... Are there any advantages in using the Prescott? I have heard from overlocking experts that the non Prescott is better because it can be pushed farther without getting too hot.

There are no definitive answers yet.

However, the Prescotts are less efficient than the Northwoods if the CPU speed is the same. This is because the Prescott pipeline is longer. Now, the longer pipeline allows the Prescotts to be clocked faster than the Northwoods.

At a certain point, strictly looking at performance, a faster clock with a longer pipeline will beat a slower clock and a shorter pipeline. I'm not sure where that line is for ffdshow.

Don't forget that the faster clock also means higher power as well.

Nice! Another definite maybe... :D That's what I have gotten all day researching this buy. Thanks anyway. And you seem like an educated guy in this field , (something I'm not because of my AMD relationship that just ended...), so let me ask you this question, then: If you were to set up a pure HTPC for ffdshow, would you go for 2.8 or 3.0, and for Prescott or non Prescott? I plan to use an Asus or an Abit mobo, both very good at overclocking (and one of those definite maybes I mentioned), and two matched sticks of 256 meg 4000 DDR memory for a total of 512.

Owen had some spec's a few pages ago, which is what I'd go for. I don't remember the specifics of the frequencies for CPU or memory, but it was based on a 2.8G Northwood P4.

Straying off topic somewhat, the reason you want to use the 2.8G P4 is strictly financial. Assuming that all the Northwoods you get right now are based on the same mask set, i.e. stepping, in a "perfect" world they would all run at whatever the maximum P4 clocks speed is across voltage and temperature. That is, however, not that case since there are variations from one batch of chips to another. Intel, and AMD, test there chips and then stamp them with a speed grade based on how fast they run. They are then "guaranteed" to run at that speed assuming your system is at the worst case limits of the spec.

But what if you know you can keep your system at the best case limits of the spec? Then your 2.8G P4 will probably be indistinguisable from a 3.X G version. This is what overclocking is all about. Lower speed grade parts have less testing done to them and/or are more plentiful, thus they are cheaper. Note that sometimes due to "good" batches there are more fast parts than slow ones. Because of PO requirements Intel/AMD will sometimes stamp a fast part with a slow number.

The other important thing to remember is that for FSB and Memory timings, 1:1 ratios are as important, and sometimes moreso, than raw frequency depending on how the memory is being accessed.

However, even my 2.4GHz P4 w/1GB of memory doesn't have the speed to do the resize at either 1440x960 (double-DVD resolution) or 1280x1024 (monitor resolution).

I do not see why you shouldn't be able to resize to 1440x960 with that 2.4g p4, which is the same as I have. I am able to do 1440x960 with Lanczos and gradual denoise at 30 no problem. Overlay yes, vmr9 no. I use sonic filters, which seem to have a lower processor load then windvd and powerdvd...at least for me.

Brian

Originally posted by usabrian
I do not see why you shouldn't be able to resize to 1440x960 with that 2.4g p4, which is the same as I have. I am able to do 1440x960 with Lanczos and gradual denoise at 30 no problem. Overlay yes, vmr9 no. I use sonic filters, which seem to have a lower processor load then windvd and powerdvd...at least for me.
I don't think I'm doing anything too strange. I'm running TheaterTek, Reclock, ffdshow, DVD Idle Pro, and Girder, all on Windows 2000 SP4. I'm not sure what to look for at this point.

-=> Mike Hanson <=-

Thanks, jvincent! I'll have a look at that!

Originally posted by boxsoft
I don't think I'm doing anything too strange. I'm running TheaterTek, Reclock, ffdshow, DVD Idle Pro, and Girder, all on Windows 2000 SP4. I'm not sure what to look for at this point.

-=> Mike Hanson <=-

I'm running a very similar setup to yours with a 2.5G P4 (and really slow 2100 RAM) and don't have any problems with Gradual Denoise(15) followed by resize (1440x960, Lanczoz).

Order matters in resize, so make sure that you have the denoise first or it will probably choke.

I'm also running XP vs 2K which may make a small difference.

Well with a P4 2.8GHz, I'm able to do:

1. Denoise3d = .5, .5, 5, HQ (very CPU intensive)
2. Unsharpen mask = 20
3. 1440x960 Lanczos resize with L&C at 1.3

Video jerks by one frame maybe twice during movies, but the pq is astounding on my 42" GWIII via DVI.

Mastiff,

I just finished the process of picking my new processor specifically to make ffdshow work better. My old processor was a 2.53Ghz with 1080Ghz RIMM and did not have hyperthreading. My new system is a 3.0Ghz prescott (faster prescotts were hard to find or expensive) overclocked to about 3.6Ghz the FSB is 246Mhz at present. I measured the total power into the prescott at 102 watts which is too high for most air cooling. This is no problem for a water cooler though. I am using a Hydocool200 to keep the CPU cool. The prescott is stable and can do Prime95 on both virtual machines just fine with the CPU temp sensor about 59 degrees. The actual chip temperature is about 13 degrees centigrade below this according to some tests done on the ABIT motherboard. The ABIT IC7-MAX3 motherboard is doing a good job and seems to have no obvious faults except that the CPU and RAM overvoltage seems inaccurate. I couldn't find the best RAM speed for this setup (3700 supposedly) and ended up getting OCZ 4000 Gold which is able to run at 1:1 and 5:4. I will experiment with 5:4 in the future to see if the FSB can be raised a little more.

If you want to go for the ultimate FSB overclock check out the 2.4Ghz c with 800Mhz FSB. Some of the games are getting 3.8Ghz out of these!

I am using ffdshow Lanczos filter to resize to 1440 by 960 with some sharpening there. If I try and use Denoise3d there is a tendency to have very short freezes on horizontal pans. It's noticeable enough to be annoying. Last night was my first attempt to use denoise3d and who knows what I might be doing wrong. If I use spline instead of Lanzcos the same (micro) freezing occurs.

Task Manager has the total CPU utilization at about 65% max but the virtual CPU where ffdshow is running is about 70 to 90% depending on what filters are running. The other CPU is running about 15%. Common belief is that hyperthreading is supposed to be very helpful but it does limit the CPU utilization (for video) to 50% with ffdshow. An experiment with hyperthreading turned off is in order as well as one with less threads running.

The Prescott tests very well with codecs and video processing and has the SSE2 instructions which may be helpful in the future. With ffdshow in constant development there is a chance that we may even get to use the SSE2 instructions sooner than later. A bigger issue is that ffdshow needs to be able to use hyperthreading to avoid that loss of about 40% of the processor power that is evident on my machine.

Hope this helps a bit.

Charlie

Anybody figure out what's the best version of FFDshow to use currently?

Originally posted by jvincent
I'm running a very similar setup to yours with a 2.5G P4 (and really slow 2100 RAM) and don't have any problems with Gradual Denoise(15) followed by resize (1440x960, Lanczoz).

Order matters in resize, so make sure that you have the denoise first or it will probably choke.
I was under the impression that the denoise should be after the resize. Regardless, I just tried your exact settings. I'm getting less jerkiness (due to the lower denoise settings), but it's still very noticeable. Also, I'm not really happy with the visual effect: there is lots of color "bleeding", like you get when a JPG is stored in "Low Quality" mode. I'll keep fiddling with it.

-=> Mike Hanson <=-

Originally posted by JoeFigueiredo
Well with a P4 2.8GHz, I'm able to do:

1. Denoise3d = .5, .5, 5, HQ (very CPU intensive)
2. Unsharpen mask = 20
3. 1440x960 Lanczos resize with L&C at 1.3

Video jerks by one frame maybe twice during movies, but the pq is astounding on my 42" GWIII via DVI.
That's interesting, as I wouldn't have expected Unsharp Mask to be used in conjunction with Resize w/L&C. I've tried this, and it had the same problem of jerkiness. On, and on, and on, and on....

-=> Mike Hanson <=-

Mastiff,

Hyper-threading was the problem. I just disabled it in the BIOS and now I can have Lanzcos 1440x960 and denoise3d with no freezing of any kind even with lots of motion. If I change to spline then picture is OK until there is a lot of motion and then a little very short freezing is visible occasionally. Under both situations Task Manager says that CPU utilization is 99 to 100%, which is believable. I might try AndyIEG's new version to see if that does the job.

Charlie

Thanks for telling me, and for the longer explanation earlier! Then I know to disable hyperthreading for now. Maybe it will be usable in a later version, as you say. As for the 2.4 I know you can clock the ... out of it, the problem is that you can fry eggs on it if you haven't got water cooling, as far as I have read today. So the best bet may be the 2.8 clocked to around 3.5 and then in a year or so to get a Prescott which then probably will be quite reasonable priced for a 3.5, that I can kick up to 4. :cool:

However, even my 2.4GHz P4 w/1GB of memory doesn't have the speed to do the resize at either 1440x960 (double-DVD resolution) or 1280x1024 (monitor resolution).

I have no trouble resizing to 1440x1152 with lanzcos on my Athlon XP 2400 running at 2600.

Try a little OC. But a "real" P4 is supposed to be better for ffdshow than a crappy Athlon XP.....

Denoise3d kills it though.

I also have a TB 2400 OC'd to about 2150 (2700xp?) running 166FSB speed.

What MB do you have, mines A7V333 (KT333) but i can't get smooth playback at that resize.

What FFDShow version?

This thread is hitting some very important topics right now. Its possible to buy motherboard/cpu combo deals real cheap right now if one went with AMD. If what some of you is reporting is true about resizing to 1440x960@72hz with an AMD2400 OC to 2600 thats great news. We need more details on how you did it. Ram and Motherboards used.
I for one just got a deal on a AMD 2500 Barton /1mb cache with ECS 333mx motherboard for $70! If that is good enough....thats a HUGE differance in price vs a full blown p4 2.8c system.

Originally posted by Mark_A_W
I have no trouble resizing to 1440x1152 with lanzcos on my Athlon XP 2400 running at 2600.

Try a little OC. But a "real" P4 is supposed to be better for ffdshow than a crappy Athlon XP.....

Denoise3d kills it though.

hehe this statement is outdated :)

try my new version for denoise3d.... and if im working hard a real AMD64 version will blow the **** out of the normal version running on a P4 :) but its a long way to go for a full working AMD64 version but i will try.

So i say it again dont buy a AMD-XP now for ffdshow u will need a SSE2 cpu AMD64 or P4 for upcoming changes in ffdshow!

Ohhhhh. Got it. Allright. I will stay focused on upgrading to a Pentium 4.
Thanks Andy.

Originally posted by AndyIEG
hehe this statement is outdated :)

try my new version for denoise3d.... and if im working hard a real AMD64 version will blow the **** out of the normal version running on a P4 :) but its a long way to go for a full working AMD64 version but i will try.


I gave your optimized denoise3D a shot.

I used Denoise 3D (0.5, 0.5, 4.44, HQ) followed by resize to 1440x960 (Lanczos).

It is definitely much better. With the 20040520 build I get a ton of dropped frames on my 2.5G P4 with PC2100 ram. Yes, I know this is a somewhat outdated CPU/MB combo.

With the optimized version my CPU goes to 80-85% and playback is pretty smooth. It still dropped a few frames, mostly when there was a big scene change. Again, given the relatively low power of my CPU/MB combo this isn't surprising.

I'm guessing that if you have a newer MB with dual channel RAM and CPU in the 2.6G+ range you should do just fine.

Other variables, Win XP, 512MB RAM, M-audio AP 24/96, AIW 9700 Pro, Cat 4.5, Reclock, DVDidle.

Originally posted by AndyIEG
hehe this statement is outdated :)

try my new version for denoise3d.... and if im working hard a real AMD64 version will blow the **** out of the normal version running on a P4 :) but its a long way to go for a full working AMD64 version but i will try.

So i say it again dont buy a AMD-XP now for ffdshow u will need a SSE2 cpu AMD64 or P4 for upcoming changes in ffdshow!

Andy,

Thanks for all your hard work. Did you fix 05-20-2004 build? I can't see horizontal scaling bug (although I tested only with high resize settings 738p and 1080p). Also it looks like green bug is gone in VMR9 (but that could have been upgrade to DX 9.0c beta - I had major upgrade weekend).

Realy looking forward for AMD-64 version. Is it going to be for XP Pro or XP 64-bit? Right now I am getting some 20-30% haircut on performance under XP-64 bit.

Mastiff,

Just a thought....

Water cooling is $211 (Hydrocool200) at newegg and is no more expensive than if a fast CPU is used. Go cheap on the CPU and use the savings to partially pay for water cooling to go fast right now with a 2.4c. Future Presotts will need the cooler anyway if they are overclocked so you can just get a new clamp and use the rest of the cooler as is.

Charlie

Originally posted by vpopovic
Andy,

Thanks for all your hard work. Did you fix 05-20-2004 build? I can't see horizontal scaling bug (although I tested only with high resize settings 738p and 1080p). Also it looks like green bug is gone in VMR9 (but that could have been upgrade to DX 9.0c beta - I had major upgrade weekend).

Realy looking forward for AMD-64 version. Is it going to be for XP Pro or XP 64-bit? Right now I am getting some 20-30% haircut on performance under XP-64 bit.

The green resize bug is gone since the 09/2003 version and since this date we have the horizontal line bug. Sorry to tell u but there was nothing changed in the code and its still in.... (all the mplayer scaler code did not change since 09/2003)
The second green bug could be fixed by your Dx 9.0c update since it was a bug in the VMR9 itself.

I will "try" to get a win-xp 64Bit version working.

PS: where u got DX 9.0c from? Or maybe u mean 9.0b?

So horizontal bug is still there. That is strange. It showed so clearly with March 25th build in Chapter 1 of LOTR RK, underwater scene, as well as in Last Samurai, the first battle (fog scenes) when Tom Cruise gets captured. Now it's gone at least from LOTR RK. I did test at same 738p. The only difference was DX 9.0c, new 60.85 beta drivers and AA and AF turned on this time on my Quadro. If not gone, it's well masked in this setup.

As far as DX 9.0c, you can get it at:

http://forums.guru3d.com/forumdisplay.php?forumid=15

The only downside is that you might have to do clean XP install once SP2 with new DX comes out (unless some DX removal tool does a perfect job which is hard to check). Looks like no clean SP2 update is possible with this ripped out version of DX 9.0c.

Thanks again for you hard work.

Just a note on Denoise3d use.

DVD color (chroma) resolution is only half that of Luma resolution so I have found that sharpening Chroma in resize does nothing but increase noise.

Also for Denoise3d I have found that I can use Chroma set to 1.0 or maybe more without any noticeable lose of detail, but noise reduction performance is noticeably improved.

So my current recommendation is to NOT sharpen Chroma and use

Denoise3d set to Luma 0.5, Chroma 1.0, Time 5.0, HQ before resize.



Andy,

I have not had time to test your new Denoise3d optimizations but I will.
Thanks for your efforts.

I tried to resize to 2160x1152 for 3x horizontal and 2x vertical PAL DVD resolution but FFDShow does not seem to accept anything over 2000.
Have you any idea why?

I am already using Denoise3d followed by Lanczos resize to 1920x1152 using Elecard decoder with overlay or VMR9 without problems. So if your Denoise3d optimizations reduce CPU load a little, 2160x1152 may be possible.

And just for the record people, I have Hyperthreading ENABLED.


Regards,

Owen

___________________________
The FFDShow resize-sharpen dude.

Originally posted by AndyIEG


try my new version for denoise3d.... and if im working hard a real AMD64 version will blow the **** out of the normal version running on a P4 :) but its a long way to go for a full working AMD64 version but i will try.

So i say it again dont buy a AMD-XP now for ffdshow u will need a SSE2 cpu AMD64 or P4 for upcoming changes in ffdshow!

ug
The Largest I would want to make my DVD's using FFDShow will be 1600x960(double the native res of my projector) and i would like to use all the other features with ffdshow as well. Since I am not using resolutions that are extremely high do i really need to dish out for the high end PC equipment? I havent bought my HTPC gear yet and I can wait a couple months for prices to drop, but right now I do not know what to buy
thanks

Charlie, I'm afraid we don't have Newegg here in Norway... Anyway, I got a good deal on a P4 Northwood 3.0, and I have bought a Zalman copper air cooler that's able to cool this thing at 3.7 gig (another norwegian with the same mobo and case that I'm using).

Andy, I hope this doesn't mean that I have made a big mistake going for a dual channel P4 instead of an AMD64? You will make a P4 version that keeps improving and will be able to compete, right? :cool:

Originally posted by Owen
Just a note on Denoise3d use.

DVD color (chroma) resolution is only half that of Luma resolution so I have found that sharpening Chroma in resize does nothing but increase noise.

Also for Denoise3d I have found that I can use Chroma set to 1.0 or maybe more without any noticeable lose of detail, but noise reduction performance is noticeably improved.

So my current recommendation is to NOT sharpen Chroma and use

Denoise3d set to Luma 0.5, Chroma 1.0, Time 5.0, HQ before resize.


Owen,

So do you suggest using the following:

Sonic decoder @ VMR9 in ZP
1. Denoise3d = .5, 1.0, 5, HQ
2. Resize = Lanczos 1440x960, with only Luma resize sharpening at 1.3, with Chroma at zero

If I can power it, do you think it's good to also use DScaler or Unsharpen mask after resize in the above setup? With Andy's new FFDShow build, I can squeeze a bit more out of my system.

Just got curious about the hyper-threading issue and ran a simple test with Prime95 which is a hyper-threading aware program that can reliably use all the CPU resources with its torture test.

With no hyper-threading and a 1024 bit FFT using 400Mb of ram it took 146 seconds to do Test1. With hyperthreading enabled but only running Prime95 on one virtual CPU it took 151 seconds to run Test1. With hyper-threading enabled and running on both virtual CPUs it took 275 seconds to run Test1 but twice the work was done.

Taking the no hyper-threading case as 100% performance this leaves the single virtual CPU hyper-threaded case at about 97% performance and the hyper-threaded two CPU case at 106% performance.

My system runs ffdshow better with hyper-threading turned off at present and I don't have a clue as to how to reverse this situation until ffdshow is hyper-threaded. At first glance the gains of hyper-threading would be quite modest anyway so my hopes that hyper-threading would be a large performance factor are dimmed a bit.

I admit to not knowing much about hyper-threading so any info or tests that would help nail this important issue down would be much appreciated.

Charlie

Originally posted by Mastiff
Charlie, I'm afraid we don't have Newegg here in Norway... Anyway, I got a good deal on a P4 Northwood 3.0, and I have bought a Zalman copper air cooler that's able to cool this thing at 3.7 gig (another norwegian with the same mobo and case that I'm using).

Andy, I hope this doesn't mean that I have made a big mistake going for a dual channel P4 instead of an AMD64? You will make a P4 version that keeps improving and will be able to compete, right? :cool:

i just finished the AMD64 techpapers and did some first try's to build a native 64bit version of ffdshow.... um forget it...

Problem 1: the 64bit compiler DONT allow inline assembler... this alone means big time recoding and rewriting of many many code lines to external asm files.

Problem 2: in 64 bit mode only SSE2 code is supported... means rewriting of all old MMX/MMX2 stuff to SSE2

Problem 3: 64bit large pointers... many little tweaks are needed

I might be concentrating on some SSE2 code now first rather than a fully optimized AMD64 version, wich means it will also run on win32 with a P4. I still have the goal to get a AMD64 version running but this will prolly take a while...
So lemme see what works best for now.

What bout SSE instead? How inferior is it to SSE2? Current Athlons, P4s and i guess 64 chips on 32 mode should all support it and gains should be seen for everyone?

(note i have no idea what im on about, just wondering)

Originally posted by Spoonfed
What bout SSE instead? How inferior is it to SSE2? Current Athlons, P4s and i guess 64 chips on 32 mode should all support it and gains should be seen for everyone?

(note i have no idea what im on about, just wondering)

SSE wont work for ffdshow, since SSE is mainly a set for fast floating point stuff, but ffdshow work with integers in 98% of the code. Means u cant load packed integers in the SSE registers but u can in SSE2. U could do some float/integer conversion but than u could also rather use MMX2 wich would perform even faster.

PS: AMD64 and P4 support SSE2 in 32bit mode mixed with mmx/mmx2 and SSE1, only on AMD64 in native 64bit mode MMX/MMX2 isnt supported anymore... dont ask me why have to read more about this crap.

oooh OK. But MMX2 is not really worth the effort (compared to SSE2). Hmmm pitty AMD don't have it in the Athlons (i guess it is a Intel thingy aint it?).

My next processor will likely be a AMD 64, and in the HTPC crowd as such high processors are only needed for FFDShow if you manage to get a version that uses this chip effectively then no doubt more will decide on this over high end P4s?

Originally posted by Spoonfed
oooh OK. But MMX2 is not really worth the effort (compared to SSE2). Hmmm pitty AMD don't have it in the Athlons (i guess it is a Intel thingy aint it?).

My next processor will likely be a AMD 64, and in the HTPC crowd as such high processors are only needed for FFDShow if you manage to get a version that uses this chip effectively then no doubt more will decide on this over high end P4s?

That was my idea to show what a AMD64 can do since im a AMD fan... but i just tested the windows 64bit edition and its far from beeing stable and perfect useable. Sure it realy works and i must say its damm fast in some areas. But many many 64bit drivers are pure crap... I guess the 64Bit revolution will only start if we have a official 64Bit windows on the market and AMD try to push the hardware devs more for stable 64bit drivers.

Btw the main reason i want do a AMD64 version is not the 64bit stuff ... that is rather useless for ffdshow, but the AMD64 add 8 more general and 8 more SSE registers wich is real big improvement. I realy wish i could use those extra regs in 32bit mode.... but u cant :(

OK, AndyIEG, so just to be perfecly shure: Even though you're an AMD fan (as I have been for four years, only ffdshow is pushing me over to the dark side) you will continue to develop for P4 as well, and I'm not doing something stupid when I'm putting together a P4 rig that will run at around 3.5-3.75 mHz? This will be my HTPC for at least two years now.

Originally posted by Mastiff
OK, AndyIEG, so just to be perfecly shure: Even though you're an AMD fan (as I have been for four years, only ffdshow is pushing me over to the dark side) you will continue to develop for P4 as well, and I'm not doing something stupid when I'm putting together a P4 rig that will run at around 3.5-3.75 mHz? This will be my HTPC for at least two years now.

Sure use the evil P4 :) i will start with some SSE2 test today and play around with compiler intristics rather than asm. I just got my answer from a Visual Studio 2005 developer (since im alpha test this version) and he told me that VS 2005 will NOT accept inline assembly for the 64bit compiler even if VS2005 goes retail... He told me i have to get all the code in a asm file and compile them with MASM or he suggest to rewrite all the asm code to compiler intristics. Im still not sure what works best for ffdshow.. lets see.

So this means there wont be a full optimized AMD64 version for quite a time and it wont matter if u use a P4 or AMD64 for new SSE2 code.

OK, thanks! I must admit that I did not understand much of what you said between "evil P4" and "it won't matter", but those told me what I needed to know. :D I am a power user, but hopeless when it comes to programming... :cool:

Originally posted by JoeFigueiredo
Owen,

So do you suggest using the following:

Sonic decoder @ VMR9 in ZP
1. Denoise3d = .5, 1.0, 5, HQ
2. Resize = Lanczos 1440x960, with only Luma resize sharpening at 1.3, with Chroma at zero

If I can power it, do you think it's good to also use DScaler or Unsharpen mask after resize in the above setup? With Andy's new FFDShow build, I can squeeze a bit more out of my system.

Owen,

Any answer on my question above?

I would say WinDVD decoder. It should be better than Sonic. Sonic's pretty old now.

Originally posted by Charles Black
Just got curious about the hyper-threading issue and ran a simple test with Prime95 which is a hyper-threading aware program that can reliably use all the CPU resources with its torture test.

With no hyper-threading and a 1024 bit FFT using 400Mb of ram it took 146 seconds to do Test1. With hyperthreading enabled but only running Prime95 on one virtual CPU it took 151 seconds to run Test1. With hyper-threading enabled and running on both virtual CPUs it took 275 seconds to run Test1 but twice the work was done.

Taking the no hyper-threading case as 100% performance this leaves the single virtual CPU hyper-threaded case at about 97% performance and the hyper-threaded two CPU case at 106% performance.

My system runs ffdshow better with hyper-threading turned off at present and I don't have a clue as to how to reverse this situation until ffdshow is hyper-threaded. At first glance the gains of hyper-threading would be quite modest anyway so my hopes that hyper-threading would be a large performance factor are dimmed a bit.

I admit to not knowing much about hyper-threading so any info or tests that would help nail this important issue down would be much appreciated.

Charlie

Prime95 is probably a really bad candidate for HT to accelerate. It is very much CPU bound with few stalls or waits for IO. HT is really still just a single CPU but it tries to keep the processor busy if one thread of execution stalls (I/O, mis-predicted branch, etc). If ffdshow has similar characteristics then it is also unlikely to gain from HT support (though multi-core or multii-processor would be a diffferent story).

Originally posted by Mastiff
I would say WinDVD decoder. It should be better than Sonic. Sonic's pretty old now.

The problems with WinDVD is:

1. Can't run VMR9 with it, unless I use AnyDVD or something like that.
2. When WinDVD freezes, video freezes completely rather than Sonic where it just skips frames and keeps going.
3. Menu selections don't work with WinDVD.

While I agree with you that WinDVD is better pq than Sonic, these little things are pretty annoying to me.

Ah. I have only used overlay so far, so I didn't know that.

Originally posted by AndyIEG
SSE wont work for ffdshow, since SSE is mainly a set for fast floating point stuff, but ffdshow work with integers in 98% of the code. Means u cant load packed integers in the SSE registers but u can in SSE2. U could do some float/integer conversion but than u could also rather use MMX2 wich would perform even faster.

PS: AMD64 and P4 support SSE2 in 32bit mode mixed with mmx/mmx2 and SSE1, only on AMD64 in native 64bit mode MMX/MMX2 isnt supported anymore... dont ask me why have to read more about this crap.

Sorry Andy, I still need a bit of clarifying re: your builds. I couldn't find my answers on either this thread or doom9. (I probably didn't look hard enough)

- The current optimizations to denoise3d are MMX2 or SSE2 specific?
- When you move to SSE2, where does that leave non SSE2 chips? Will they fall back to MMX2 optimizations? (which would still be superior to your pre-optimized code I assume)
- Do these builds include the new lanczos resizer or is that in the future?
- You are just optimizing the operation of denoise3d with assembly code, not changing the way it works right?

Thanks for working on this stuff!

- The current optimizations to denoise3d are MMX2 or SSE2 specific?
A: MMX2

- When you move to SSE2, where does that leave non SSE2 chips? Will they fall back to MMX2 optimizations? (which would still be superior to your pre-optimized code I assume)
A: i think best would be to build a normal mmx/2 version and a SSE2 only version.

- Do these builds include the new lanczos resizer or is that in the future?
A: ? there is no "new" lanczos" resizer atm. If there is a "new" resizer it will prolly be a SSE2 tuned version, since the actual use MMX2.

- You are just optimizing the operation of denoise3d with assembly code, not changing the way it works right?
A: Yes i dont change the algos itself. They work the same way than before.

Thank you, and thanks again for working on the ffdshow project. Seems like your lycos site is down right now... are there any other mirrors?

Originally posted by JoeFigueiredo
Owen,

So do you suggest using the following:

Sonic decoder @ VMR9 in ZP
1. Denoise3d = .5, 1.0, 5, HQ
2. Resize = Lanczos 1440x960, with only Luma resize sharpening at 1.3, with Chroma at zero

If I can power it, do you think it's good to also use DScaler or Unsharpen mask after resize in the above setup? With Andy's new FFDShow build, I can squeeze a bit more out of my system.


I dont recomend Sonic as it does not work well in software mode, as it must with FFDShow. It gives a soft and noisy image from my memmory.
It's been a long time since I used Sonic
I prefer WinDVD with VMR9.
Or for best performance, I use Elecard with Overlay on my system.

1. Denoise3d = .5, 1.0, 5, HQ

Yes

2. Resize = Lanczos 1440x960, with only Luma resize sharpening at 1.3, with Chroma at zero

Yes, but the amount of sharpening is up to you and depends on your display and preferance. 1440x960 is for NTSC. PAL should be 1440x1152.

"do you think it's good to also use DScaler or Unsharpen mask after resize"

Maybe, it's up to you. Again it depends on your display and preferance.

Regards,

Owen

___________________________
The FFDShow resize-sharpen dude.

Owen,

I managed to get FFDShow (fairly new version), Elecard build 2510 (i found it eventually) and Zoomplayer (WMV version) working well for DVDs and im quite happy with the result. (wish i had more grunt for FFDShow but).

Is there any thing special need to get the menu items to "highlight" on selection, i've had no luck here. Menu navigator works fine with the mouse, but they do not highlight, which makes using the remote hard.

Is there anything special for PAL that may be causing the issue?

I had to installed DVD Region Free to get Elecard working due to that pop error thingy one gets.

Spoonfed,

For the record, here's my config that wherein the navigator highlighting works:

Elecard 2.00.0.3611
Zoomplayer4 beta3
Elecard(No CSS) selected
DVD Idle Pro for DeCSS
DVD Navigator Filter:Microsoft DVD Navigator
Video Renderer: VMR9 Windowed
Radeon 9500 non-pro/catalyst 4.5
Directx 9.0b

With this config, everything works great for DVD playback.

Truckchase

Hmmm im using Overlay (and older elecard build) which could perhaps be the menu highlight issue? I might try VMR9, ive never had it look as good as overlay but.

Perhaps it could also be that im using DVD Region free and you are using DVD Idle Pro?

Soooooo close, there always some little issue :) Oh also we are PAL here, don't know if that would matter

I did a quick test of Andy's optomized Denoise3d and found it to be at least 30% faster.

Great work Andy. This will make Denoise3d useful to many more people.

How much performance increase can we expect from SSE2 optimization?

Regards

Owen

Originally posted by Spoonfed
Owen,

I managed to get FFDShow (fairly new version), Elecard build 2510 (i found it eventually) and Zoomplayer (WMV version) working well for DVDs and im quite happy with the result. (wish i had more grunt for FFDShow but).

Is there any thing special need to get the menu items to "highlight" on selection, i've had no luck here. Menu navigator works fine with the mouse, but they do not highlight, which makes using the remote hard.

Is there anything special for PAL that may be causing the issue?

I had to installed DVD Region Free to get Elecard working due to that pop error thingy one gets.

I'm not sure Spoonfed, I rip almost all of my DVD's so they have no menus and I control with a mouse so I never notice the problem.
I'll have to check, but I think I do not get highlighting in menus

Regards,

Owen

Originally posted by Owen
I did a quick test of Andy's optomized Denoise3d and found it to be at least 30% faster.

Great work Andy. This will make Denoise3d useful to many more people.

How much performance increase can we expect from SSE2 optimization?

Regards

Owen

hard to tell, its kinda hard to explain but lets say in mmx mode u can work with 2x 32bit integers per instruction, in SSE2 u can work with 4.
That would make SSE2 2 times faster "theoretical"
The problem is that the cpu can do up to 3 MMX commands per cpu cycle, while in SSE2 mode it can do 1 and 1/2 per cycle, but it also depends on the instructions.
Atm i try to write a MMX2 version using only compiler intrinsics, wich is as fast as the asm handcoded one. I want first check this since for a 64bit release i can only work with compiler intrinsics and i dont want do all the work again...

oki u want a number.... lets say +20-40%

With the optimized denoise3d version, I can almost run denoise3d AFTER resize at 1440x960 with Elecard.

What do you think is better:

Running denoise3d after resize at 1280x720
OR
Running denoise3d before resize at 1440x960

Joe, what kind of horsepower does your system have? Just curious.

P4 2.8C 800FSB o/c 10% to 3.08
512 DDR PC3200 400MHz
Asus P4P800se mobo
Nvidia Aopen fx5900xt

Running Elecard 2510, Reclock 1.4, Windvd audio, and of course, FFDShow

You can click on my profile to see the complete package.

Great, thanks. Makes me think I might be able to get up to Denoise3d after 1440x960 once I fully push my 2.8c to 3.5.

-MP

Absolutely, and probably even run it after resizing (assuming you are doing nothing else.

Any ideas on my original question?

I'd say resize to 1280 and run it after... but that's assuming your display device is native 1280, which I think it is?

My display device is a 42" GWIII, so the native is quite irregular, however my desktop resolution is 1280x720

Ohh 30% difference, i'll have to have a try. Athlon TB's have MMX2 im pretty sure?

Good work. Keep up the AMD 64 work, i think that will be my next processor within the next 6 months (unless intel bring out something super, which will probably be too expensive anyways) :).

Spoonfed, don't knock Intel. I have been using AMD exclusively for 3-4 years now, but I wanted the best for ffdshow, so I defected to the dark side. I got my P4c 3.0 today, with an Abit IC-7 Max 3 board. With the stock cooler I have 3.6 gHz stable right now, running for four hours with 100 % load for stress testing! Tomorrow I'm getting a Zalman 7000 copper cooler, and I recon I will be able to reach 3.75 pretty easily! Right now the Abit mobo reports 54 degrees C, but they are notoriuous for showing 7-10 degrees too much.

Andy, are memory timings important for ffdshow? I have the system running at an FSB of 240, so the memory speed would be 480. I'm using DDR42000 memory right now, so the timings are pretty bad, 2,5-4-4-7. But is that important for ffdshow? If it is I'll send the memory back (I have the right to do so for two weeks now - that's norwegian law when you buy something online) and get 3700 memory that will get me down to 2-3-3-7, at least. :cool:

Mastiff,

Way to go! Better keep track of the CPU temp with something like Motherboard Monitor 5 though since if things get way too hot the CPU will gradually slow down... and you don't want that! :)

My 3.0e is maxed at 248 FSB - above that Prime95 starts to have round-off errors. 100% CPU utilization gives 63c on the die with water cooling, your CPU will have less power to dissipate so water cooling would probably be a waste of money. Hope you will let us know what the numbers are when you get the over-clocking done.

Good luck!

Charlie

With them PC specs both with and without over clocking what resize can you go up to?

Currently I only go to 1024 x 768 (my output res is 1024 x 576)
Thats at around 84% CPU with a gradual denoise, slight unsharp, deinterlace etc.

@Mastiff im not 100% sure but on my test's the memory timing dont help much for ffdshow. Sure there is a effect but take into account that the reads/writes comes mainly from L1/L2 cache, ffdshow dont work that much with uncachable direct memory access. So even if its a memory intensive application 90% of the cpu usage goes to integer calc's from the cpu.

I realy hope that the new sse2 code will help to run ffdshow for everyone without buying expansive hardware. The goal is to be able to resize (2x dvd res) + sharpen + denoise3d + color conversion on an cheap AMD64 2800+ with normal pc2100 (266 ddr ram) without speed problems.

Btw 90% of the ffdshow user just resize + sharp + denoise + ??? What else filter is most used?

Thanks, Charlie! From what I have heard (and I have been a staple at OCForums.com the last week or so researching this) the problem is that the Prescott's for now is less efficient than Northwood, and also it has a problem with high temperatures. I use MBM5, yes. And when the computer goes online I will set E-mail warning to around 55 degrees. Right now I'm seeing gradually lower temps on both load and idle, today (after about 20 hours use with stock cooling and Arctic Silver5) I have 38 idle and 55 full load, probably because the silver is setting. Too bad I'm going to take it out again in a few hours, when I get the Zalman... Anyway, I have two blowholes and extra fans in the top, plus the special Abit Max3 "OTES" heat removal (a duct and a fan taking away heat from capcitors and whatnots above the CPU) so I'm not too worried about running too hot. And of course you'll hear how far I get! :D

Spoonfed, I haven't gotten around to that yet, and I probably won't until the beginning of next week. I will take my sweet time setting up this PC, so it's 100 % stable and ready for prime time when I put it in my HT. That means images every time I have done anything substantial (like Windows Update, SP1 and especially antivirus, firewall and ATI drivers. And then the HTPC related stuff, like Girder... Wonder if I get any work done at all this "work weekend" (my wife and kids are at her parents' for two and a half weeks, and I always work extra then).

And thanks, Andy! I will test DDR3700, but if I can't get that stable at the speed I'm expecting I will stick to the 4200 then. And I use the dScaler sharpen filter all the time. :cool:

I tried you latest version. While at the resize i do my setup can't 3D Noise (CPU spikes to 100%, then stuter, then smooth (90%) then stuter (100%), tis very weird) (i'll have to stick with gradual) DPU usage has still dropped buy about 5% or so at least.

Every little bit helps, thanks :)

Originally posted by AndyIEG

Btw 90% of the ffdshow user just resize + sharp + denoise + ??? What else filter is most used?

I think there are also a fair number of Dscaler sharpen users.

Originally posted by AndyIEG
@Mastiff im not 100% sure but on my test's the memory timing dont help much for ffdshow. Sure there is a effect but take into account that the reads/writes comes mainly from L1/L2 cache, ffdshow dont work that much with uncachable direct memory access. So even if its a memory intensive application 90% of the cpu usage goes to integer calc's from the cpu.

Andy,

If cache is the key performance factor for FFDShow, than it looks like P4 EE would be by far the best ($$$ and sense aside) FFDShow CPU. With its 2MB L3 cache its FFDShow performance would probably be off the charts. Now it is clear why my FX-51 performs better than Athlon 64 3400+. It is not the dual chanel memory, it is 1MB of on-die cache. Similarly, my FX-51's 1MB cache is why it is able to catch up (with respect to FFDShow performance) with P4s running at 3.6 Ghz while clocked at only 2.5 Ghz.

On a different matter, I downloaded FFDShow version from the link provided in your signature (that one registers properly in Forceware Multimedia 3.0 beta, as opposed to all other April/may builds), and this one shows input from decoder at 29.97 fps, compared to March 25th that was showing 24 fps. I still use the same decoder and same settings, so I was wondering which one is right? I will likely have the opportunity to test my HTPC on one of Sony's new 1080p digital projectors (Qualia), but the beast only takes 1080p 24 fps. I thought Forceware Multimedia 3.0 beta would output 24 fps to FFDShow. If it only does 29.97, I think it would be difficult for video card's drivers to downconvert to 23.97 fps without audio/video synch issues.

Vlad: Couldn't Reclock help somehow with sync issues? I'm out of my league in terms of programming, but perhaps this is ogo's next chance to shine.

Originally posted by vpopovic
Andy,

If cache is the key performance factor for FFDShow, than it looks like P4 EE would be by far the best ($$$ and sense aside) FFDShow CPU. With its 2MB L3 cache its FFDShow performance would probably be off the charts. Now it is clear why my FX-51 performs better than Athlon 64 3400+. It is not the dual chanel memory, it is 1MB of on-die cache. Similarly, my FX-51's 1MB cache is why it is able to catch up (with respect to FFDShow performance) with P4s running at 3.6 Ghz while clocked at only 2.5 Ghz.


The part about how your 2.5ghz amd runs equal to the 3.6ghz P4 is not because of cache but because of a multitude of reasons. The whole architecture of the chips are very different, and Ghz no longer means what it used to. In fact, Intel will completely do away with GHZ tagging in their product lines either with mature prescotts or intels next chip. Its very confusing already to decide between 2.8 and 2.8A and Intel realizes this.

I have found some benchmark tests for recent chips but none of them use FFDshow. What type of app is it in the first place? is it an encoder etc?
here are the benchmark tests of video programs which include Windows media encoder, Main Concept MPEG encoder, XMPEG, Pinnacle STudio 9, Windows Movie Maker and more.
Just go through the pages to find the rest of the benchmarks in the easy to read chart form.


http://www6.tomshardware.com/cpu/20040322/prescott-15.html
http://www6.tomshardware.com/cpu/20031223/cpu-guide-17.html

Originally posted by JBlacklow
Vlad: Couldn't Reclock help somehow with sync issues? I'm out of my league in terms of programming, but perhaps this is ogo's next chance to shine.

We'll see. The only way to try is to have DVI connection to 1080p 24 fps device.

Originally posted by Hawkson
The part about how your 2.5ghz amd runs equal to the 3.6ghz P4 is not because of cache but because of a multitude of reasons. The whole architecture of the chips are very different, and Ghz no longer means what it used to. In fact, Intel will completely do away with GHZ tagging in their product lines either with mature prescotts or intels next chip. Its very confusing already to decide between 2.8 and 2.8A and Intel realizes this.

I have found some benchmark tests for recent chips but none of them use FFDshow. What type of app is it in the first place? is it an encoder etc?
here are the benchmark tests of video programs which include Windows media encoder, Main Concept MPEG encoder, XMPEG, Pinnacle STudio 9, Windows Movie Maker and more.
Just go through the pages to find the rest of the benchmarks in the easy to read chart form.


http://www6.tomshardware.com/cpu/20040322/prescott-15.html
http://www6.tomshardware.com/cpu/20031223/cpu-guide-17.html

Unfortunately none of these video encoding tests reflect ffdshow performance since ffdshow isn't an encoder, but rather a directshow filter/post-processor. If anything, it's closer to a decoder.

Originally posted by Hawkson
[B]The part about how your 2.5ghz amd runs equal to the 3.6ghz P4 is not because of cache but because of a multitude of reasons. The whole architecture of the chips are very different, and Ghz no longer means what it used to. In fact, Intel will completely do away with GHZ tagging in their product lines either with mature prescotts or intels next chip. Its very confusing already to decide between 2.8 and 2.8A and Intel realizes this.


You are right. Chips are completely different. I did not explain all of my thought process to get to the conclusion. So here is how I think it goes.

If you look at some posts where AMD 64 3400+ users reported their FFDShow results, you will see theat they are very shy of what 3.6 Ghz P4 and my FX-51 can do. FX-51 does not run FFDShow much faster in dual chanel memory setup vs. single chanel based on Andy's input, and based on my experience (Ops, I ran for about a week without noticing that my second RAM stick was not seated properly, so the system was running only one 512 MB stick in single chanel mode). So difference in FFDShow performance between two AMD's can not be attributable to dual vs. single chanel RAM. Otherwise, FX-51 (@2.5 Ghz) is quite similar to AMD 64 3400+, except, of course 1 MB on die cache. So it is the additional cache that saves the day. Here FX-51 is beefier than P4. To determine that Ghz realy matters is quite simple. At stock settings of 2.2 Ghz my FX-51 gets about 15% haircut in performance. This is consistent with % of overclock. It runs marginally better when I add to the clock speed (that ups frequency of HT, PCI and AGP), rather than overclocking by increasing the multiplier (pure CPU speed).

If the "on-boad" cache is the key here, then Prescot P4:s should make the day since they have 1 MB cache (L2) on the chip - something in between a Nortwood P4 with 512 KB (L2) and P4EE with 2 MB (L3).

If this is true, then I will "go after" (that is buy) a Prescot P4 next week........

Originally posted by Lars
If the "on-boad" cache is the key here, then Prescot P4:s should make the day since they have 1 MB cache (L2) on the chip - something in between a Nortwood P4 with 512 KB (L2) and P4EE with 2 MB (L3).

If this is true, then I will "go after" (that is buy) a Prescot P4 next week........

You have to be careful, while it may in fact be the cache, the Prescotts cores are different than the Northwoods so the extra cache may not make a difference.

In order to tell, you'd have to test a Prescott and Nortwood running at the same frequency in the same machine to see if it made a difference.

I don't really follow the CPU architecture battles anymore, are the P4 caches equally split between instruction and data? Intruction cache would make a big difference by and of itself but given that the dataflow in ffdshow probably involves a lot of loads and stores I'm not sure that the data cache will help that much.

P4s don't have a conventional instruction cache. They use a trache cache instead. In any case, I really wish there would be a review with benchmarks catering to the ffdshow audience, but I guess we're a niche market :(

By no means I thought that my conclusion would be right. It is merely a deductive conclusion from the information we exchange through this forum. I have no better guess, but guess is still a guess.

In other forums they have shootouts, perhaps it is time to start that here. No better way to compare video cards, decoders, CPUs etc. then by having a bunch of people together looking at the same thing.

True. Why not simply start a thread at first. People post their exact PC specs. Motherboard/chipset. Ram (brand/speed/bios etc), Processor and overclocking amount (and how ie muliplier etc), Graphics card used, etc etc.

This will at least give a rough indication of what people looking to upgrade with FFDShow in mind what can be achieved.

As i only have a lowely 2400+ (at the time (1 year ago) i only tested for HD which it did easy, new nothing about FFDShow etc), so obvioulsy im probably scrapping the bottom of the barrel here :) hehe

There was a thread a while back that did exactly what you say. Two members posted (me included) and it just sank to the bottom of the forum like a rock. I really don't need it as I have as good of a CPU as it gets (short of P4 EE), but I am always ready to share.

Originally posted by vpopovic
I really don't need it as I have as good of a CPU as it gets (short of P4 EE), but I am always ready to share.


You can ship the "shared" CPU to ........................... :) hehe j/k

Yeah i see your point. Though perhaps now more members are aware of FFDShow than ever? I have had my HTPC for about 1 year, yet have been using FFDShow much less than that and use to skip past much FFDShow talk (Stupid me)

hehe oki i will cleanup some things here

First the ffdshow code is a "patched" code means many code lines comes from other projects and various sources and are patched together. Next is that many of the asm functions are coded for mmx and to run on an old pentium in mind. Wich slows down some of the routines on dif. cpu's. For now u cant simple say memory or cache size is the key. The key is how "happy" the cpu is with some lines of the code.
So for example in many asm routines the result of an mmx register is direct copied into an normal 32bit register. On an P4 we dont get a penalty but on an AMD cpu we get a huge penalty by doing this. Next example, on a P4 a mmx to mmx register copy takes 2-3 times longer than on an AMD. There are many more examples and so some penaltys for the one cpu weight off some for the other...

I can only say its realy hard to handcode a asm routine wich works "perfect" on all the diff. cpu's... Thats also the reason why i will prolly exclusive work on a P4/AMD64 version and strip all the old mmx code and try to replace them. The main adv. by doing this is i can concentrate on the AMD64 and P4 architecture and dont need care about old P2/P3 or athlons. Its also easyer to later port the code to a 64bit version.

maybe this helps a bit ;)

That sounds like a very good idea! My P4 3.0C @ 3.75 has been running for around 20 hours with full load, so I call that stable. Which probably means that it'll be able to run whatever ffdshow throws at it in the forseable future. Looking forward to hook it up to the Barco and start testing! :cool:

That sounds like a terrible idea!:eek: I plan to run my AthlonXP into the dirt before I spring for a K8 (when prices are down to my level).

But I know I'm being a Luddite, so I'll just ask which earlier version is best for a K7?

Keh,

The best version for smoothness, and lower CPU usage i have tried on my Athlon 2400 (over clocked to 2166, 2700 i think?) is Andy's latest version.

All this techie stuff is starting to make my head spin. All I want is to make my HTPC run ffdshow as efficiently as possible. Here's what I have:

P4 Northwood 3.06GHz w/ Hyperthreading
Asus P4P8X 533MHz FSB, Dual Channel
2 x 256MB PC2100 266MHz

I am currently running @ 3.4 GHz (overclocked 10%) and am at the edge of ffdshow performance.

With my MoBo, if I replace the ram with PC3200(400MHz), the FSB will run at 800MHz. Will this boost performance enough to justify the cost?

I'm really confused about all of this overclocking stuff. How should I run this combo? I'm running stck fans now, but am plannning on buying a Zalman cpu fan. How do I set that up to prevent damage.

Sorry for these questions, but I really have no desire to learn this stuff, I'd just like someone to tell me how to do it! :-)

Thanks,
Phil

Nice little FAQ, good work.

Have been a fan of AviSynth for awhile now, use it to feed material into CCE. So when I stumpled accross this I could'nt pass up the chance to see what all this was about.

Probelm was that my old HPC config was fine for playing DVD's but lacked the processing power for FFDshow. So ordered a new 2.8E GHz chip, overclocked to 3.4 (with new motherboard and RAM) and popped that into my Antec Overture.

Following instructions from the posted link I was up and running in no time and was watching LOTR ROTK on my Pioneer SD-643HD5.

I used the sonic DVD codecs along with Zoomplayer. I don't size my screen up to 1776 x 1000 as I don't think my set likes it. 1080 x 480 is fine.

Will I stop using my Pioneer DV-47A player? Doubtful. Still having a hard time grasping the concept that this type of configuration will blow away a $500.00 DVD player.

But I encode alot of MPEG2 material and wanted to get away from burning DVD's hence the HTPC. This endeavor only makes that process more enjoyable.

Cheers

Greg Y

pmd918
Looks like your motherboard is the bottleneck. It only supports 533mhz. You need one that supports 800mhz which is what Hyperthreading needs.

HT doesn't "need" a 800MHz FSB. It just so happens the P4C CPUs are usually paired with i865PE/i875P chipset based motherboards. The P4 3.06 is a 533MHz FSB CPU that supports HT, normally running on i854PE platforms.

Hmm last post didn't post. Strange.

Goi I didn't know that CPU was 533mhz. So that motherboard that he has is fine then. Just need to match your new ram to it now. Not sure if that increase in performe will be enough for ffdshow thou.

Have a question. What is that refresh rate for 72hz? 71.928 ?

Also, ANyone finally settle on the perfect or close settings with Denoise3D?
Your settings would be most usefull.
Using a 9200se with 1440x960@7hz. Marquee 8500 crt projector.

I posted this on the 1st


QUOTE Owen

"Just a note on Denoise3d use.

DVD color (chroma) resolution is only half that of Luma resolution so I have found that sharpening Chroma in resize does nothing but increase noise.

Also for Denoise3d I have found that I can use Chroma set to 1.0 or maybe more without any noticeable lose of detail, but noise reduction performance is noticeably improved.

So my current recommendation is to NOT sharpen Chroma and use

Denoise3d set to Luma 0.5, Chroma 1.0, Time 5.0, HQ before resize."

Owen

Just as a FYI, the latest FFdshow beta (200440601b) works wonder with NTSC DVDs using TheaterTek on my lowly Celeron 2.0 overclocked at 2.66 GHz (corresponding to a 533MHz FSB).

For the first time, I can apply Denoise3d with Owen's settings AND resize to 1440x960. The picture is 'perfect' when the framerate is around 24 fps. With menus and 30fps animation, there are some dropped frames, but that is not a major issue in my book.

I must not be far from the limits of my system, as PAL DVDs overwhelm my machine when I try to resize to 1440x1152... Maybe, with some of Andy's magic I'll be able to enjoy them too!

Eiffel

pmd918, you should turn off hyperthreading if it isn't already turned off. I read earlier in this thread that it actually takes away processing power from ffdshow. As for the Zalman (I assume you mean one of the 7000's, I use the pure copper one and I have overclocked my 3.0 to 3.75 100 % stable and without any heat problems with that fan only running on 7V (which is what it does if you use the Fanmate that comes with it without turning the potmeter down any)) the only thing you should watch out for is that you need better thermal compound than the cheap (deleted) that Zalman gives you. Arctic Silver 5 is the stuff to get. With that you should be pretty safe. As long as the CPU voltage (VCore) isn't bumped too much up, it will be OK. But do bump the memory voltage up to 1.80, if not you probably won't get it stable. Also often it helps to increase the AGP voltage even though you're not overclocking the graphics card. This is only a small fraction of the knowledge I was able to pick up this week when I asked a lot of newbie questions and searched a lot at an overclocking forum. :cool:

Mastiff. Can you tell me what the setting looks like in BIOS to bump up the AGP voltage? I saw a setting I didnt know what it was with strange settings. 2 digits with one being a letter of the alphabet. A thru M i think.

Originally posted by Eiffel
Just as a FYI, the latest FFdshow beta (200440601b)

Where did you get this?

Originally posted by stylinlp
Mastiff. Can you tell me what the setting looks like in BIOS to bump up the AGP voltage? I saw a setting I didnt know what it was with strange settings. 2 digits with one being a letter of the alphabet. A thru M i think.


In Bios there is a setting called AGP Driving Value
the options are:

Set at DA right now. But the range goes from DA to D9
All the way up thru E9
Thats alot of choices for something I have no clue what it means and Im A+ certified technician lol

Originally posted by blackmax2k1
Where did you get this?

from http://mitglied.lycos.de/ieggei2/ffdshow/

Originally posted by stylinlp
In Bios there is a setting called AGP Driving Value
the options are:

Set at DA right now. But the range goes from DA to D9
All the way up thru E9
Thats alot of choices for something I have no clue what it means and Im A+ certified technician lol

I don't think that's it, but then again I don't know what motherboard you're using. Most of them have it on the same page as the one where you're able to view/adjust the rest of the CPU/memory parameters such as FSB, multiplier, Vmem, Vcore, FSB/mem ratio, AGP divider, etc...

ok I have a report. Not good, sigh. :(

I have an AMD 1000 w/ 256megs PC133 ram. Win XP, Randeon 9200se. Cat 4.5, Theater Tek, ffdshow 20040602b Marquee 8500 crt projector.

I just enabled Denoise3D with Owens prescribed settings. Thats the only filter I turned on. Sure enough I was spiking 100% CPU most the time. ALot of stuttering. So much I couldn't enjoy the movie. But what I did notice was a HUGE differance in image quality. That filter practically made the noise in the back ground disappear. Wish I could have kept it on.

I am referancing to the movie "Girl with the Pearl Earing" 1st scene, the back wall behind her. Lots of little squiggly worms all over it. Enabled that filter and all that noise was gone. Perfectly smooth wall. Nice.

I do have a AMD 1800 laying arond collecting dust but didn't want to waste time re-installing windows just to get one filter to work. When (hopefully) soon I will upgrade to a P4 2.8c.

Originally posted by stylinlp
ok I have a report. Not good, sigh. :(

I have an AMD 1000 w/ 256megs PC133 ram. Win XP, Randeon 9200se. Cat 4.5, Theater Tek, ffdshow 20040602b Marquee 8500 crt projector.

I just enabled Denoise3D with Owens prescribed settings. Thats the only filter I turned on. Sure enough I was spiking 100% CPU most the time. ALot of stuttering. So much I couldn't enjoy the movie. But what I did notice was a HUGE differance in image quality. That filter practically made the noise in the back ground disappear. Wish I could have kept it on.

I am referancing to the movie "Girl with the Pearl Earing" 1st scene, the back wall behind her. Lots of little squiggly worms all over it. Enabled that filter and all that noise was gone. Perfectly smooth wall. Nice.

I do have a AMD 1800 laying arond collecting dust but didn't want to waste time re-installing windows just to get one filter to work. When (hopefully) soon I will upgrade to a P4 2.8c.

sorry to hear this my optimisations arnt perfect for the mmx2 version... i nearly finished the AMD64/P4 SSE2 version wich is ALOT faster than the mmx... maybe if i have time i review the mmx2 code again and see what i can do for more speed.

Wow, Andy's ffdshow 040601c version is quite a bit faster on my P4C than the previous 0508 official build I was using earlier. Great work.

Andy.....Keep up the work. The latest version you released, is just working wonders on my set up.


Athlon 64 3200
ZP/Sonic Decoders


I am able to run the Denoise 3d at th esettings suggested and it is a much easier load on my unit then previous version.

Originally posted by AndyIEG
sorry to hear this my optimisations arnt perfect for the mmx2 version... i nearly finished the AMD64/P4 SSE2 version wich is ALOT faster than the mmx... maybe if i have time i review the mmx2 code again and see what i can do for more speed.

Andy,

Is optimization only for denoise, or resize as well? Thanks again for the great work.

Originally posted by stylinlp
In Bios there is a setting called AGP Driving Value
the options are:

Set at DA right now. But the range goes from DA to D9
All the way up thru E9
Thats alot of choices for something I have no clue what it means and Im A+ certified technician lol

that setting is more for stability issues and not for performance improvement. this is an explanation of the option from Adrian's Rojak Pot Bios optmization guide. the site is located at http://www.rojakpot.com/

AGP Driving Value

Common Options : 00 to FF (Hex numbers), 00h to FFh

Quick Review

This BIOS feature will only be activated if you set the AGP Driving Control BIOS feature to Manual. It determines the overall drive strength of the AGP bus.

The drive strength is represented by Hex values from 00 to FF (0 to 255 in decimal). The higher the drive strength, the greater the compensation for the motherboard's impedance on the AGP bus.

In conjunction with AGP Drive Strength and AGP Drive Strength P Ctrl, this function is used to bypass AGP dynamic compensation in cases where the auto-compensation circuitry cannot provide adequate compensation. If you are using an AGP card built around the NVIDIA GeForce 2 line of GPUs, then it is recommended that you put AGP Driving Control into Manual mode and set AGP Driving Value to EA (234). For other cards, please check with the manufacturer if your card requires the AGP driving strength to be manually set.

Incidentally, increasing the AGP drive strength will not improve the performance of the AGP bus. It is not a performance enhancing feature so you shouldn't increase the AGP drive strength unless you need to.

Originally posted by vpopovic
Andy,

Is optimization only for denoise, or resize as well? Thanks again for the great work.

The next release will have a SSE2 version for the denoise3d and dscaler sharpen filter and some internal routines are also rewritten for sse2. Mainly the stuff wich is easy to convert and easy to understand.

Im than working my way torward the resizer routines and color conversion.... but those routines will take some time since they arnt that easy to convert.

Andy, stop it, you lureing me into a new rig (likely AMD64 if you optimisations are good). hmmmm :)

What would be better a normal Athlon64 chip with the standard cache of an FX/operton chip with the extra cache? Or no difference? (price is substantially different).

I found a great deal on a AMD 3.0 cpu with 333 fsb. Matched with PC3200 512megs ram.
Would this be enough for Andy's new version coming out to Resize at 1440x960@72hz ?

Oh one more question. Sounds like a dumb question but maybe I just didnt see this before. Funny how someone can just assume things.
If ffdshow Resizes to 1440x960 do we still need Powerstrip to create that resolution? Seems redundant now that I think about it.

Originally posted by stylinlp

Oh one more question. Sounds like a dumb question but maybe I just didnt see this before. Funny how someone can just assume things.
If ffdshow Resizes to 1440x960 do we still need Powerstrip to create that resolution? Seems redundant now that I think about it.

They're somewhat orthogonal issues. ffdshow is controlling what goes IN to the video card while Powerstrip is controlling what comes OUT of the video card.

Originally posted by stylinlp
I found a great deal on a AMD 3.0 cpu with 333 fsb. Matched with PC3200 512megs ram.
Would this be enough for Andy's new version coming out to Resize at 1440x960@72hz ?

AMD 3.0 ?? Whats this huh

Dont panic dudes all u need is a SSE2 capable cpu thats all. If im realy very very lazy than maybe a 64bit version will pop up some day. But for now i dont think most of u would use windows xp 64bit even if a full 64bit version would be out. (this would be a AMD64 only 64bit version btw)

So a normal AMD64 or P4 will do the work. If u can spare some extra money for a bigger cache this also helps since im working with many manual cache prefetches. But im not realy sure how much faster the routines work with larger cache... maybe they arnt at all.

RIght now on Egghead.com you can get AMD Barton CPU's at a good price.
Barton AMD's have a 512k Cache. I was looking at the following choices.
AMD XP 2500 Barton, 512k cache, 333fsb
or
AMD Athlon XP 3000+ "Barton", 333MHz FSB, 512K Cache Processor - OEM
Model: AMD Athlon XP 3000+
Core: Barton
Operating Frequency: 2.16GHz
FSB: 333MHz
Cache: L1/64K+64K; L2/512K
Voltage: 1.65V
Process: 0.13Micron
Socket: Socket A
Multimedia Instruction: MMX, SSE, 3DNOW!, 3DNOW!+
Packaging: OEM(Processor Only)

AMD Athlon 64 3000+, 512KB L2 Cache, 64-bit Processor - OEM
MMX, SSE, SSE2, 3DNOW!, 3DNOW!+
Socket: Socket 754

Athlon 64 is much more expensive. +$100

If u realy want to use ffdshow with all the hardcore settings and huge resize go for a SSE2 cpu's, i wont work on the older mmx code and SSE1 is useless for ffdshow. The price diff. from a AMD xp to a AMD64 is not that high just get a cheap board (i have a cheap asrock for 60€ and it works) im also using my old DDR266mhz ram wich also is enuff for ffdshow. All u need is SSE2 ram speed dont matter, expansive FSB bla bla dont matter. The instruction set and the Mhz of teh CPU matter most!

Oh.................
ok :)
hmmm....I see what what your saying that your going to have both AMD and Pentium cpu's optimized. So the choices are AMD 64 or Pent4 Hyperthreading CPU's. Now I know I can shop around for the cheapest combo. The Pentium cpu/mb combo with that ram was going over budget. Glad we dont have to suffer than soon. Thank you once agian Andy :)

May I ask. Which ffdshow version optimizes SSE? I might be stuck with an AMD Althlon 3000 Barton.

jvincent
Thank you for that answer to PowerStrip :)
Couldnt figure out why ffdshow OSD was showing In 720x480 and Out 720x480 with just Denoise3D enabled.

As he has said (a few times now) SSE instructions do not work with what FFDShow needs to do!!!

Is there a way to have FFDShow only work with DVDs?

It kills even my P4 3.2 when trying to work its magic on a 1080I HD clip.

Has anyone run into this issue and have a good solution?

I guess you are using ZP? If so in the "customised media" tab take FFShow out of the filter list for mpeg2.

However this will disable it for SD also. (not for DVD as it is separate section).

It is a good question, i would like ZP to have it enabled for SD but not for HD

Just setup autoload conditions in ffdshow based off the dimensions of the movie and enable/disable whatever you want for that media type. (or disable everything)

Sorry I haven't chimed in earlier - went away for the weekend.

Thanks to all who replied to my post earlier in this thread.

But my main question wasn't answered. The manual for my mother board specifically states that it will run at 800MHz with PC3200 ram. The "normal" FSB speed is 533MHz with the PC2100 ram that I'm using now.

So, my main question stands - will it be worth while to upgrade my ram and run at 800MHz? I assume the answer is yes, but I just don't know enough about this stuff to know whether there isn't some other bottleneck that I should worry about.

I'm not going to totally revamp my HTPC to boost performance, I just thought that if I could spend a few bucks on ram to get a significant improvement, I'd do it. If not, no big deal.

Thanks,
Phil

Originally posted by pmd918

So, my main question stands - will it be worth while to upgrade my ram and run at 800MHz? I assume the answer is yes, but I just don't know enough about this stuff to know whether there isn't some other bottleneck that I should worry about.



Based on the work Andy is doing, good stuff BTW Andy, I would hold off on any CPU upgrades for the next little while.

Once he's finished with his SSE2 upgrades you may have enough horsepower to do what you want.

jvincent,

Thanks for the reply. I just re-read the latest couple of dozen posts, and I see what you mean.

Begs another question, though. I am running the 20040418 version of ffdshow, seems like I should try one of the newer versions? Or maybe just wait, as you suggest, until the upgrades are complete?

Thanks again,
Phil

Then if thats what Andy says I would assume this cpu below would do it then.

Intel Pentium 4/ 2.4A GHz 533MHz FSB, Prescott Core, 1MB L2 Cache - Retail (limit 10 per customer) $125


Model: Intel Pentium 4
Core: Prescott
Operating Frequency: 2.4GHz
FSB: 533MHz
Cache: L1/12K+16K, L2/1MB
Voltage: 1.25-1.525V
Process: 0.09Micron
Socket: Socket 478
Multimedia Instruction: MMX, SSE, SSE2, SSE3
Warranty: 3-year MFG
Packaging: Retail (with Heatsink/Fan)

The Pentium4 2.8c hyperthreading would cost at least $60 more.

But it can be overclocked a lot further :)

Originally posted by pmd918
Begs another question, though. I am running the 20040418 version of ffdshow, seems like I should try one of the newer versions? Or maybe just wait, as you suggest, until the upgrades are complete?


It's worth trying the current test version, if for no other reason to give Andy some feedback.

Something must be amiss in my system. After going through the majority of this forum, I would have thought my HTPC had enough power to handle FFDSHOW to its full potential. I've already posted my issues with video sources, but I can deal with those. This weekend, I was trying some more tweaking, and found that I can't go beyond:

FFDshow 5/20/2004
1. Denoise3D = 0.5, 0.5, 5, & HQ
2. Resize = 1280x720 with Lanczos@10 and L&C each at 1.5

My projector's natice resolution is 1280x720. Many people suggested resizing to 1440x960, but I get skips when I try that. Still other suggest resizing first, but again I get skips.

My machine:
MB: Intel D865PERL w/865PE chipset
CPU: Intel P4 2.8CGHz with HT on
Mem: 512 Meg (Kingston DDRAM 256MX2 DDR400 C3 PC3200 K RTL)
Audio: M-Audio AP2496 with v27 drivers
DVD Player: TheaterTek 1.5.64
Video: ATI Radeon 9600 (non-pro) with CAT 4.5 drivers
OS: WinXP Pro SP1 with all latest patches

Other info:
- All Windows sounds are off
- Most non-essential services are off
- I only use this machine to run Media Center 10 and TheaterTek.

Several people on this forum with similar setups seem to be able to resize higher without a problem. The only thing I can think of is maybe my MB or RAM isn't up to the task. During DVD playback, my CPU is barely at 50%, so there seems to be plenty of CPU headroom available.

With my setup, would you expect that I could resize higher or even first? Are there some other tweaks I could try to squeeze a little more performance out of my system? Note, I prefer to avoid overclocking. Absolute stability and quiet is a must for this machine.

Thanks!

You can try the new test version Andy has linked in his sig.

Originally posted by madpoet
You can try the new test version Andy has linked in his sig.

If u have a P4 or AMD64 pls wait. I will release a new P4/AMD64 (SSE2) version soon.
So far the new denoise3d SSE2 version, new SSE2 dscaler sharpen, yv12 color conversion are done. Im just doing some cache prefetch adjustment and test's and maybe rewrite the yuy2 conversion too.

Originally posted by godavego
During DVD playback, my CPU is barely at 50%, so there seems to be plenty of CPU headroom available.


Turn off HT and you'll see that your CPU use is really much closer to 100%. That second vertual CPU really throws the measurements off for video playback. Stock 2.8 may be a little light for doing everything you are trying, most posts have it OC'd over 3 to get reasonable performance. Certainly try the test version or wait for a full SSE2 optimized though if you'd rather not OC. With all of the optimizations AndyIEG has been doing you may be able to run everything you want at stock speeds.

Anyone know if DVD Region Free works like AnyDVD does so that you can use Windvd in ZP for VMR9?

Aspect ratio...

I've read all of this thread, trying to understand how to setup TheaterTek & FFDshow to get proper aspect ratio.

I am using TT1.5.64, FFDshow 2004-06-01a, Catalyst 4.5 driving component video at 1280 x 720, Windows XP Pro SP1.

Without FFDShow, I get appropriate full screen width on a movie like 'Master & Commander'. As soon as I enable post-processing, the movie is displayed within a four sided window. It appears to be the correct AR, just reduced from full screen width.

In FFDshow, I am resizing to 1440x960, Lanczos, parameter 10, with 'no aspect ratio correction' selected.

In TT, the aspect ratio shows as 16x9 as expected.

What am I setting incorrectly?

After installing Andy's ffdshow test version ffdshow-20040601c.exe WinDVD5 crashes (FWMM 3.0 too) going back to ffdshow-20040520.exe fixes it, any ideas? does it only work with cinemaster codec?
Thx

Originally posted by JavierS
After installing Andy's ffdshow test version ffdshow-20040601c.exe WinDVD5 crashes (FWMM 3.0 too) going back to ffdshow-20040520.exe fixes it, any ideas? does it only work with cinemaster codec?
Thx

Crash happens also for me, trying to fix this fast and removing the "c" version till this is fixed.

godavego, turn off hypterthreading. And you may try to overclock that one a bit as well - if you got a desent cooler you should be able to go up to at least 3.2. I got my 3.0 up to 3.6 with the boxed cooler.

Originally posted by JavierS
After installing Andy's ffdshow test version ffdshow-20040601c.exe WinDVD5 crashes (FWMM 3.0 too) going back to ffdshow-20040520.exe fixes it, any ideas? does it only work with cinemaster codec?
Thx

Found the bug, the bug is in the actual cvs checkout not in my code. As small fix delete the following reg entry using "regedit"

HKEY_CLASSES_ROOT\CLSID\{00000000-0000-0000-0000-000000000000}

will be fixed in the new version

Wow, talk about getting put in my place! I fired up the finally finished install and decided to try for double PAL resolution (1440x1152 - isn't that double Pal?), denoise 3D HQ and dScaler sharpen. My CPU is totally locked on 100 % both in Windows and on the ffdshow OSD and I get dropped frames. This is on overlay, I haven't even tried VMR. Is this normal? If it is, I may have to go down a notch on the resize, but I actually thought that a 2.75 P4 should be able to handle this. The version is from the 5. of April. Maybe that's not a good version?

Oh, and the ffdshow OSD shows same input and output size. Is that normal?

So you are reszing and then applying Denoise3d? You definately need the new version in Andy's link, and even then you might not be able to handle it.

No, no, I'm using Denoise 3d first. I wrote it wrong. I use both denoise and dscaler sharpen before resize.

Ahhh... either way, try the new version. He improved Denoise3d's CPU use.

Torr, one issue could be the OSD, as it exhausts resources very quickly! Have you tried disabling it?

Eiffel, that does not help. It's really maxed out. But I will try the new test version. I'll report back in a while! :cool:

Edit: No, I won't. Andy's obviously doing something right now, the test version isn't there.

He's finishing up optimizing for SSE2, so that AMD64s and P4s should see a gain.

OK, that sounds good! I'll wait until he posts again here to say that it's done. :cool:

Mystery solved. I was on Spline resize instead of Lancoz. With Lancoz I can resize to 1440x1152, with DeNoise3d and dScaler sharpen before it. And man, what a difference that made in the picture as well!" This is going to be fun! My wife will think that I have bought a new projector while she was away visiting her parents. Hey, wait a minute! I could do that! She'll never see the difference in my BG808 and a new BG808S as long as it's hidden under the plattform in the HT! ;)

I do have a very slight jitter on extreme pans (like when Gandalf comes to Bilbo's house in the beginning of FotR, but I'm sure the new denoise and resize algorithms (if they're algorithms - it's just a buzzword I have found somewhere...) will fix that!

Incredible! I can even run VMR9 with minimal jitter with double DVD resolution and those other two filters! I AM SUPERMAN!!! Up, up, and awayyyy!!!!! :D

OK, down to earth again. There isn't much difference between VMR9 and overlay on my Radeon 9800, but there is one thing: On overlay I have a thin, dark green stripe on the right side, visible only in very light scenes. Does anybody know what that may be? :cool:

nearly finished the new version's

BTW i found a bug in the latest nvidia drivers(i use 61.12) wich caused 30% higher cpu usage for my system.

It happend as i was using vmr9 and tryed to set brighness and other stuff with zoomplayer or render color control. After i changed some settings the driver save those in the reg and the slow down is in every mode i used since (overlay/vmr9). I was wondering why i lost 30% cpu power... was thinking it was some crazy stuff in my new code..

Finaly i found the fix... -> video->settings->extended->gforce pannel->overlay controls->RESTORE DEFAULTS

after this all runs smooth again :)

Good! I'll be waiting. As for the Nvidia bug: Yet another reason to go to ATI. Their bugs usually only affects gamers... ;)

my new versions are up, a normal and a new P4 and AMD64 only version
feel free to test them

here is the changelog:

2004-07-06 22:00 Andy2222

* SSE2 optimized denoise3d filter (only the "HQ" version)
* new SSE2 optimized FLT_Sharpness_SSE2.dll included with ffdshow (dscaler sharpen)
* fixed memcpy ref. to the faster SSE2 version

2004-07-06 9:00 Andy2222

* fix for WinDVD CLSID crash
* set xvid as default IDCT
* disabled mlib AC3 decoder as default decoder

2004-06-06 8:00 Andy2222

* small fix in denoise3d filter (again)

2004-06-05 12:00 Andy2222

* small fix in denoise3d filter


PS: in the future i will prolly only release new P4/AMD64 only versions, the P2/3/Athlon... version will only come from Athos like alway's