DXVA compatibility for software players thread [Archive]

tetsuo55

21st May 2008, 13:00

This is a work in progress:

This is a thread about all the problems regarding playback of files with DXVA enabled, results from the discussion here regarding encoding settings should be discussed here (http://forum.doom9.org/showthread.php?t=132924).

Section 1: DXVA implementations:
Section 2: Software players supporing DXVA
Section 3: Videocards supporting DXVA
Section 4: Getting DXVA to work
Section 5: Sample file's for testing purposes

Section 1: DXVA implementations

There are basically 2 kinds of DXVA:
1) Hardware assisted decoding, the videocard offloads some parts of the decoding process from the software decoder, this is very complex to code and you basically need help from videocard developers to be able to include this in your program.
2) Full bitstream decoding, the videocard does everything, the player only has to feed the video stream to the videocard, pretty simple and straightforward, you can assume that this is the way forward for all videocards and software-players

Section 2: Software players supporing DXVA

1) PowerDVD ultra, this program supports both types of DXVA and has support for all codecs (WMV9.MPEG2,VC1,H264)
2) WinDVD, this program supports both types of DXVA and has support for all codecs (WMV9.MPEG2,VC1,H264)
3) MPC-HC, this program supports only the 2nd type, full bitstream decoding, and does so for all codecs (WMV9.MPEG2,VC1,H264), however it decodes WMV9 externally with a combined 1+2 method, also this is the only program that is free and open-source

Section 3: Videocards supporting DXVA
ATI
Bitstream:

All cards starting with HD

Assisted (supported codecs are behind the card's name):

NVIDIA
Bitstream:

Assisted (supported codecs are behind the card's name):

Section 4: Getting DXVA to workto be done
Section 5: Sample file's for testing purposesto be done

There is a special MPC build compiled only for testing videocard abilities, Warning this player can and will crash your computer so keep this in mind when testing, get it here (http://www.etud.insa-toulouse.fr/~mvelten/bordel/mpc-hc/mplayerc_r518_TEST.zip)

I need help with the following:
* Links to sample files. For each file technical details (levels, max ref frames, basically the exact encoding settings+output from avinaptic or a similar program) should be included.
* A list of graphics card models that are (somewhat) compatible. If a card has some limitations with regard to the type of files that it supports, then that info should also be included. to test your videocard use the following program: http://bluesky23.hp.infoseek.co.jp/index.html#DXVAChecker
* For each sample file we need to keep a list of testresults of which cards are able to play it and which are not.

MarcioAB

21st May 2008, 13:28

nVidia 8600 GTS (with up to date driver) here and the main point is the Ref Frames>4 limitation. Anything else I tried so far works fine.
To test, I just pick ANY source (from low res to hi res) and re-encode using x264 varying any parms in x264. If I respect the max Ref Frames = 4 anything else works.
By work I mean, play very nice on MPC-HomeCinema with DXVA turned on and CPU usage stays on 2% or less

Because the nVidia driver is Unified, maybe this Ref Frames limitation could be true for all nVidia cards that has DXVA capability.

Any way, if this is a limitation, may we need to live with that and ask everybody else to encode respecting max Ref frames = 4.
By the way, MeGUI is profile oriented and there are several profiles with came ready to use for DXVA and all of them already respect the max Ref Frames = 4.

MatMaul

21st May 2008, 14:03

nVidia 8600 GTS (with up to date driver) here and the main point is the Ref Frames>4 limitation.
have you tried those files with the cyberlink decoder ? It is weird to be limited by a specific number of frames and not by a width*height*nb_ref value.

I'm looking for a sample which is known to have problems with nvidia (macroblocks for example) and the mpc decoder but works fine with cyberlink decoder.

tetsuo55

21st May 2008, 14:09

MarcioAB, please use this program and paste the results in a new post

http://bluesky23.hp.infoseek.co.jp/index.html#DXVAChecker

Kado

21st May 2008, 16:37

MatMaul

21st May 2008, 16:43

Also check the videos from this post (http://forum.doom9.org/showthread.php?p=1139516#post1139516), for me only "x3_sample.mkv" plays without video corruption.
no prob with cyberlink with those 3 mkvs ?

Kado

21st May 2008, 16:56

None. Just rechecked them.

tetsuo55

21st May 2008, 17:20

Kado and Matmaul
please use this program and paste the results in a new post

http://bluesky23.hp.infoseek.co.jp/i...ml#DXVAChecker

don't forget that PowerDVD has assisted DXVA too, which could be working around the problems with those file's

rack04

21st May 2008, 17:24

Casimir666

21st May 2008, 21:32

3) MPC-HC, this program supports only the 2nd type, full bitstream decoding, and does so for all codecs (WMV9.MPEG2,VC1,H264), however it decodes WMV9 externally with a combined 1+2 method, also this is the only program that is free and open-source
To be accurate, mpc decode only VC1 and H264 with DXVA (not Mpeg2 and WM9) ;)

Kado

21st May 2008, 21:39

@tetsuo55
I just use the cyberlink's decoder filter and the profile is "Bitstream". I also have another file that does not play with mpc internal dxva but goes fine with cyberlink's filter using dxva but is too large to put online, it has high bitrate and the cpu consumption is about 2 to 5% meaning that the gpu is decoding everything.
http://pwp.netcabo.pt/kado/dxva_mpc.jpg

Kado

21st May 2008, 22:06

@casimir666
Can you compile a recent version of MPCVideoDec.ax? :thanks:

Casimir666

21st May 2008, 22:26

@casimir666
Can you compile a recent version of MPCVideoDec.ax? :thanks:

Here it is
http://www.zshare.net/download/1237146982e56028/

MarcioAB

21st May 2008, 23:55

MarcioAB, please use this program and paste the results in a new post

http://bluesky23.hp.infoseek.co.jp/index.html#DXVAChecker

NVIDIA GeForce 8600 GTS
ModeMPEG2_C: DXVA1, NV12, 720x480 / 1280x720 / 1920x1080
ModeMPEG2_D: DXVA1, NV12, 720x480 / 1280x720 / 1920x1080
ModeVC1_IDCT: DXVA1, NV12, 720x480 / 1280x720 / 1920x1080
ModeWMV9_IDCT: DXVA1, NV12, 720x480 / 1280x720 / 1920x1080
ModeVC1_MoComp: DXVA1, NV12, 720x480 / 1280x720 / 1920x1080
ModeWMV9_MoComp: DXVA1, NV12, 720x480 / 1280x720 / 1920x1080
ModeVC1_PostProc: DXVA1, NV12, 720x480 / 1280x720 / 1920x1080
ModeWMV9_PostProc: DXVA1, NV12, 720x480 / 1280x720 / 1920x1080
1720AC81-9D1B-4F63-9A37-4A88483D0B87: DXVA1, NV12, 720x480 / 1280x720 / 1920x1080
ModeH264_VLD_NoFGT: DXVA1, NV12, 720x480 / 1280x720 / 1920x1080

gngn

22nd May 2008, 01:04

pretty much the same results here as MarcioAB. using MPC-HC r518 with internal filter and Nvidia 175.16 (latest) drivers on a 8600GTS, i can only play H264 with Ref Frames<=4 (edited, initialy said ">4" by mistake) , but without seeking.
everything above 4 will result in a black screen, with no playback, no matter the Output Renderer. if i put MPC from windowed into fullscreen or the other way around i get to see 1 frame perfectly displayed.

as for the cyberlink decoder, i dont get it. it will play alot more diferent files than the MPC-HC internal decoder, but not all. i have a file that the internal decoder playes , ReFrames : 3, but not the cyberlink decoder. sample (http://rapidshare.com/files/116651035/_Rumbel__Soul_Eater_-_01__H264-HD_-001.mkv)

NVIDIA GeForce 8600 GTS
ModeMPEG2_C: DXVA1, NV12, 720x480 / 1280x720 / 1920x1080
ModeMPEG2_D: DXVA1, NV12, 720x480 / 1280x720 / 1920x1080
ModeMPEG2_IDCT: DXVA2, NV12, 720x480 / 1280x720 / 1920x1080
ModeVC1_IDCT: DXVA1/2, NV12, 720x480 / 1280x720 / 1920x1080
ModeWMV9_IDCT: DXVA1/2, NV12, 720x480 / 1280x720 / 1920x1080
ModeVC1_MoComp: DXVA1/2, NV12, 720x480 / 1280x720 / 1920x1080
ModeWMV9_MoComp: DXVA1/2, NV12, 720x480 / 1280x720 / 1920x1080
ModeVC1_PostProc: DXVA1/2, NV12, 720x480 / 1280x720 / 1920x1080
ModeWMV9_PostProc: DXVA1/2, NV12, 720x480 / 1280x720 / 1920x1080
1720AC81-9D1B-4F63-9A37-4A88483D0B87: DXVA1, NV12, 720x480 / 1280x720 / 1920x1080
ModeH264_VLD_NoFGT: DXVA1, NV12, 720x480 / 1280x720 / 1920x1080

MarcioAB

22nd May 2008, 01:10

@MatMaul
This video (http://rapidshare.com/files/113536801/Suzumiya_Haruhi_no_Yuuutsu_-_C4_-_Special_Ending_-__Hitsuji__e569a2b7_.mkv.html) opens with cyberlink decoder(dxva) but does not open with mpc internal using dxva.
Also check the videos from this post (http://forum.doom9.org/showthread.php?p=1139516#post1139516), for me only "x3_sample.mkv" plays without video corruption.
Vista SP1 + EVR pre-SP1 + mpc rev 518 + 9800gtx 175.63.

Before play this "Suzumiya" video I checked with MediaInfo that it was encode with 11 Ref Frames so my expectation was a black screen, but to my surprise it show up.
Checking CPU usage I see around 15%, so despite show up with DXVA on, it's clear it's not using hardware acceleration.

I will check the other videos on the suggested thread.

MarcioAB

22nd May 2008, 02:00

i can only play H264 with Ref Frames>4 , but without seeking.

Hum, so you CAN play something with 5 Ref Frames or more. I can not.
Regard the seeking (well observed): I can not use seeking at all with DXVA on any of my encodes.

everything above 3 will result in a black screen, with no playback, no matter the Output Renderer.

Not follow that: 4 Ref Frames or more = black screen ?
Over here I have 5 Ref Frames or more = black screen.

i have a file that the internal decoder playes , ReFrames : 3, but not the cyberlink decoder.

I can play such 3 Ref Frames file in DXVA mode but not able to seek.

MarcioAB

22nd May 2008, 02:13

Also check the videos from this post (http://forum.doom9.org/showthread.php?p=1139516#post1139516), for me only "x3_sample.mkv" plays without video corruption.
Vista SP1 + EVR pre-SP1 + mpc rev 518 + 9800gtx 175.63.

None of them work for me (the first 2 are 8 Ref Frames each and last 2 are 9 Ref Frames each).

gngn

22nd May 2008, 02:40

oh, sorry for the mixup. i meant Ref Frames < 4 (i did copy-paste from your previous post :D). i dont have any at all with 4 to test so i can only say that those files with Ref Frames 2 or 3 will work on my XP with the DXVA of MPC-HC. i did reencode one file to Ref Frames=4 from originaly 8, and it works too (but only from the second attempt, at the first attempt after i played any other file it will show a black screen, maybe due to my lacking knowlegde of encoding? my reencode (http://rapidshare.com/files/116664289/_SS-Eclipse__Kyouran_Kazoku_Nikki_-_01__1280x720_h264_-001_RF_4.mkv)).
and like i said, the only file that the internal decoder of MPC-HC does open, but not the cyberlink decoder is that sample i pointed in my earlier post.

none of those 4 samples that Kado mentiones work for me. and with cyberlink i get a lot of frames lost

MarcioAB

22nd May 2008, 03:07

Ok. And your 4 Ref Frames reencode works fine (as all the others, seek does not work ... still need to discover why).

Now Kado (with a 9800 GTX) said one of those 4 samples worked for him.

We are all using the same driver. I'm on XP, Kado on Vista, not sure what OS you are.

Considering OS is not filtering the bitstream for non-complaint (resulting the black screen), I think the driver is also not filtering and so the bitstream is going directly to the hardware resulting a "hardware black screen failure". It seems 8600 (NV12) hardware can not handle "5 Ref Frames or more" and 9800 (NV24) is "in-between" to handle. Maybe future 10800 (or whatever the name) will be able to handle consistently "5 Ref Frames or more". It seems support for hardware acceleration is a "work-in-progress" for nVidia. Any way, even with "4 Ref Frames or less" capable-hardware, it's already a pleasure to watch smooth 1080p H264 videos with MPC-HomeCinema.

gngn

22nd May 2008, 03:39

well it's certain that 8600 cannot decode anything over 4 Ref Frames perfect. with the cyberlink decoder it will work in some cases (still huge frameloss), like the 5 samples that Kado provided, that have 8 resp 9 Ref Frames (the anime one 11), but those are not true 720p. and seeking does work with the cyberlink decoder.
is there a chance that some of the problems regarding the geforce 8000series will be worked out by some new nvidia drivers?

tetsuo55

22nd May 2008, 07:03

as for the cyberlink decoder, i dont get it. it will play alot more diferent files than the MPC-HC internal decoder, but not all. i have a file that the internal decoder playes , ReFrames : 3, but not the cyberlink decoder. sample (http://rapidshare.com/files/116651035/_Rumbel__Soul_Eater_-_01__H264-HD_-001.mkv)

This file works for me too in MPC-HC.

Okay let me get this straight people. For NVIDIA:

MPC-HC DXVA will only work with 4 ref frames or less, seeking will only work with 3 ref frames or less

PowerDVD DXVA will work with any number of ref frames within the level limit and seeking always works when the video does

Assuming what i just said is correct, it has already been proven that powerDVd is also using bitstream so that means that MPC-HC either has a bug with the bitstream feed or that NVIDIA has a bug with Purevideo and PowerDVD has found a workaround, either way casimir will have to look into this then

Any way, if this is a limitation, may we need to live with that and ask everybody else to encode respecting max Ref frames = 4.

This is not a limitation its actualy a bug, all DXVA devices will play the streams within their level limits (ATI, XBOX360,PS3,PSP,Nokia,Standalone Bluray, HDDVD and h264 Players) it's only nvidia and it seems only with MPC-HC that is causing problems

smok3

22nd May 2008, 08:47

tetsuo55

22nd May 2008, 09:42

so what card should i get? (currently nvidia 6600)
it must;
a. run easily at 2x(1920x1200) at least
b. have two independent outputs for connecting a 2nd screen (think LCD/plazma TV)
c. must fully support DXVA
d. is not ATI

No videocard exists that can do what you want

The only option you have is 2 ATI cards, but they cannot be in crossfire mode

Kado

22nd May 2008, 13:58

Before play this "Suzumiya" video I checked with MediaInfo that it was encode with 11 Ref Frames so my expectation was a black screen, but to my surprise it show up.
Checking CPU usage I see around 15%, so despite show up with DXVA on, it's clear it's not using hardware acceleration.

Now Kado (with a 9800 GTX) said one of those 4 samples worked for him.

There seems to be some confusion regarding my results, all the 4 sample files play using the internal mpc dxva decoder but only one displays a perfect picture, the other have corruption/blockiness, it is a bug for nvidia cards that casimir is aware of (http://forum.doom9.org/showthread.php?p=1139503#post1139503) and with cyberlink decoder they all play fine.
The Suzumiya Haruhi video does not play with internal dxva, the video is not even processed, only audio is played.
Now regarding that cyberlink is "not using hardware acceleration" with the Suzumiya Haruhi video:
http://pwp.netcabo.pt/kado/sh_dxva.png
Those results were obtained with dxva checker v1.5.1.0, let the video running until the end, so it's obvious it's using dxva by looking at the results.
CPU is a Pentium D 930 @ 4.2GHz.

MarcioAB

22nd May 2008, 15:31

MPC-HC DXVA will only work with 4 ref frames or less, seeking will only work with 3 ref frames or less

Well, not here. Even for 3 Ref Frames seeking is not working.

PowerDVD DXVA will work with any number of ref frames within the level limit and seeking always works when the video does.
Ok. Downloading Cyberlink PowerDVD trial to check on my rig.

This is not a limitation its actualy a bug ...
Ok. A MPC-HC bug.

Looking at the 11 Ref Frames "Suzumya" sample @Kado indicated, it's interesting to notice that MPC-HC somehow recognize it will not play with DXVA-on and so automatically plays it with DXVA-off. This does not happens with the other samples ( with Ref Frames between 5 and 9). On those cases (5-9 Ref Frames), it look likes that MPC-HC just throw the stream directly to the (non-capable) hardware, that will freeze (black screen). But now assuming it will work with Cyberlink PowerDVD (on so it's a bug of MPC-HC) ... need to check. (PowerDVD download is so sloooow).

MarcioAB

22nd May 2008, 15:42

The Suzumiya Haruhi video does not play with internal dxva, the video is not even processed, only audio is played.Over here (8600 GTS) it is processed but "not using hardware accelaration" even despite MPC-HC is with DXVA-on.
(something like MPC-HC recognize it will not work using hardware and so automatically turned it off).

Now regarding that cyberlink is "not using hardware acceleration" with the Suzumiya Haruhi videoI did not try Suzumiya on Cyberlink PowerDVD because I do not have the trial version yet (downlooonging ... something wrong with that site).

Those results were obtained with dxva checker v1.5.1.0, let the video running until the end, so it's obvious it's using dxva by looking at the results.Not sure how to use DXVAChecker 1.5.0.1. Instructions seems to be in Japanese. I just double-click in the executable and see the written information.

Kado

22nd May 2008, 19:13

Over here (8600 GTS) it is processed but "not using hardware accelaration" even despite MPC-HC is with DXVA-on.
(something like MPC-HC recognize it will not work using hardware and so automatically turned it off).

I disable the ffmpeg in the decoder to force videos to play via DXVA and if they are not dxva playable there's no image if external filters are blocked.

Not sure how to use DXVAChecker 1.5.0.1. Instructions seems to be in Japanese. I just double-click in the executable and see the written information.

Update to the latest version (v1.5.1.0), then open the executable and drag and drop a video over the window, then all filters capable of opening will show on the bottom of the window with some other info, if you can't see it stretch vertically the window. Also you can play the video using evr and vmr9, with or without dxva (if available), just right click on the part with the info on the filter you want to use then select the options, the modes in red are the dxva capable.
Example picture (http://pwp.netcabo.pt/kado/dxva_checker.jpg)

I did not try Suzumiya on Cyberlink PowerDVD because I do not have the trial version yet (downlooonging ... something wrong with that site).

I'm not using the whole power dvd, just the decoder filter with mpc hc, to see the "statistics" and the "decimation" options hold CTRL then open the filter properties.
Again a pic. (http://pwp.netcabo.pt/kado/pdvdf.jpg)

P.S.=> I really like to use images for exemplifying things :)

pirlouy

23rd May 2008, 00:10

Nice thread.

I will be a bit offtopic, but I just want to clarify some things.
I have a 8800GT. I've tried DXVA offered by last build of winDVD (ok, some weeks ago) but it wasn't perfect, video was jerky.

DXVA from MPC-HC is not compatible with my video card (if I try to enable, it causes problems).

But what I want to say, it's the fact that DXVA is not an obligation. I see sometimes people ready to buy a videocard in order to have DXVA support but you should really buy a better processor. I use a E8400 (ok, overclocked to 4Hz) and I can watch all these videos (http://forum.doom9.org/showthread.php?p=1139516#post1139516) without problems with post-processing at maximum. Task monitor shows me less than 20% with theKMplayer.

So, in the future, if you want good image quality, I recommend getting a better processor instead of finding and buying a videocard DXVA compatible. DXVA is difficult to use (depends on ATI/nvidia/Cyberlink/etc. which are not very cooperative) and prevents from using post-processing and others filters.

MarcioAB

23rd May 2008, 04:12

I think the seeking problem with DXVA on MPC-HC (at least with 8600 GTS) is related to hit the I-frames. To auto start after a seek, the seek must hit an I-frame. If I encode with all I-frames (of course) there is no seeking problem at all because any seek will fall over an I-frame.
If I encode with I-frame at each 5, 4, 3 or 2 frames I have less and less difficult to auto start after a seeking because I increase my chances the seek fall over an I-frame.

So, my question is: Is that possible that with DXVA-on, MPC-HC reposition an user seek as if the user have selected the seek over the next I-frame ?

I checked how seek performs without DXVA, in an encode with -scenecut 100 and -keyint 999: It starts far from the point where the seek was indicated. So in software controlled play, it seems some kind of this "move to next I-frame" do exist. We must have it now when the play is hardware controlled.

Thank you.

MarcioAB

23rd May 2008, 04:29

DXVA from MPC-HC is not compatible with my video card (if I try to enable, it causes problems).Hum ... did you try encodes with max 4 Ref Frames ? (check with MediaInfo or anything else).

But what I want to say, it's the fact that DXVA is not an obligation.Sure. But I think the point is that for movies with high bandwidth (or low compression) full HD (1080p) H264 with lots of B-frames, with much scene change plus large subtitles etc ... will somehow hit the CPU. Of course a powerful CPU should handle that. I think primarily the graphics cards are for gaming but why not else to accelerate such movies.

.. and prevents from using post-processing and others filters.Yes, that is a good point. I also mention that for the new DivX H264 decoder that is coming.
http://forum.doom9.org/showpost.php?p=1139962&postcount=119

GTPVHD

23rd May 2008, 04:46

Err no, hardware accelerated decoding has better deinterlacing and noise reduction, no matter how fast the CPU you have, you cannot do deinterlacing and noise reduction without frames being dropped. Nvidia achieved max points in HD HQV benchmark for HD decode quality. Post processing and other filters can never achieve the same image quality at real-time speeds as hardware decoding.

http://www.elitebastards.com/cms/index.php?option=com_content&task=view&id=559&Itemid=29&limit=1&limitstart=3

Even with the NVIDIA driver at its default setting it beats out the ATI board, mainly on account to superior performance when it comes to removing jaggies (with all three lines in HQV's test proving to be free of jaggies with the GeForce 8800 GT compared to just two with the Radeon HD 3870 X2) and more notably in the film resolution loss test, where the ATI card fails to deliver the required image quality with some artifacts appearing during the test. With noise reduction enabled in the NVIDIA driver, it does indeed score a perfect one hundred points in this test - A most impressive feat.

Casimir666

23rd May 2008, 07:08

ohropax

23rd May 2008, 13:06

mpc-hc decoder seek to the I-Frame already. Seeking problem with nVidia came from an error return by the driver. This error arrive when too much data are send to the nVidia accelerator, this have been fixed in latest beta release on the other thread.

I cannot confirm seeking being fixed in r518. I have a low-resolution ref=3 movie that cannot be seeked in MPC-HC but can in PowerDVD.

Is there a list of proper sample videos that we all can use for testing? If yes, could you (casimir666?) post links or update the initial posting accordingly? Or should everyone that is willing to assist only post results from his private collection instead?

A question in order to build up a representative collection of samples for all forum users: Is it problematic from a copyright point-of-view if we cut our files so that only the first seconds (containing a title-independent intro/trailer) remain? Often enough these snippets are enough to test whether playback works or not.

arfster

23rd May 2008, 13:36

Err no, hardware accelerated decoding has better deinterlacing and noise reduction,....

Deinterlacing/denoising and all other postprocessing are done in the shaders after the decoding stage, it's a different process. You can do software decoding and still have hardware deinterlacing/denoising - just use a decoder that outputs NV12. For example, the ATI mpeg decoder, coreavc, ffdshow (remember and set bob output flag for deinterlacing).

MarcioAB

23rd May 2008, 13:51

I cannot confirm seeking being fixed in r518.Me too. r518 or r518_TEST. Seek does not work with Ref Frames=1,2,3,4, with any number of B-frames, I turned off everything ... the only thing that touch the seek-issue is the I-frames. If the click is right over an I-frame it does work.

@Casimir666: When you say MPC-HC "seek to the I-frame already", can I think that despite MPC gets the real position where the user click, it will ignore that and reposition such click as if the user had really clicked on the next I-frame ?

Thank you.

MarcioAB

23rd May 2008, 14:01

A question in order to build up a representative collection of samples for all forum users: Is it problematic from a copyright point-of-view if we cut our files so that only the first seconds (containing a title-independent intro/trailer) remain? Often enough these snippets are enough to test whether playback works or not.

Another idea, assuming anyone can re-encode, is just post the key parms to allow the creation of a "same/similar" sample but with local content. I think in this hardware-accelerated field the issue is more on the people's hardware differences that can not be reproduced.

metalmania

23rd May 2008, 15:36

The perfect combination for DXVA compatibility is Nvidia 8X00 + PowerDVD CL264dec.ax (Version 1.99.0.1405) + Forceware 163.75

I have a 8500GT running in Windows XP with 163.75 driver and CL264dec.ax version 1.99.0.1405. With MPC-HC, I can play any x264 file, 720p or 1080p, avi or mkv, with DXVA enabled, no stutter, no 20 fps bug, no black screen, no purple screen, no split screen... However if I change anything in that combination, the thing goes to hell.

So I keep this combo as the video playing combo, and I have a 8800GT system with the new forceware driver to play games. I wish the DXVA solution could be found, so I don't have to stick to the old perfect combo.

Casimir666

23rd May 2008, 19:14

@Casimir666: When you say MPC-HC "seek to the I-frame already", can I think that despite MPC gets the real position where the user click, it will ignore that and reposition such click as if the user had really clicked on the next I-frame ?Thank you.

On seek, parser should position to the first keyframe before the seek point. Decoder decode all frames from I frame to seek point, but drop them (should not be rendered).

gngn

23rd May 2008, 21:30

well , seeking does not work for me either. if i reencode a video and set the "Keyframe Interval = 1", the seeking will work everytime, at any position. but videoquality drops like hell :D

parameters: Encoding settings : cabac=1 / ref=4 / deblock=1:-2:-1 / analyse=0x3:0x113 / me=umh / subme=6 / me-prepass=0 / brdo=1 / mixed_ref=1 / me_range=16 / chroma_me=1 / trellis=1 / 8x8dct=1 / cqm=0 / deadzone=21,11 / chroma_qp_offset=0 / threads=3 / nr=0 / decimate=1 / mbaff=0 / fgo=0 / bframes=3 / b_pyramid=1 / b_adapt=1 / b_bias=0 / direct=3 / wpredb=1 / bime=1 / keyint=1 / keyint_min=1 / scenecut=40(pre) / rc=2pass / bitrate=1000 / ratetol=1.0 / rceq='blurCplx^(1-qComp)' / qcomp=1.00 / qpmin=10 / qpmax=51 / qpstep=4 / cplxblur=20.0 / qblur=0.5 / ip_ratio=1.40 / pb_ratio=1.30 / aq=2:1.00

MarcioAB

23rd May 2008, 21:56

keyint=1 / keyint_min=1 ... yes.

I also confirm that MPC-hc + Cyberlink decoder + nVidia = no "seek" problem.

Obs: to me, nVidia = 8600 GTS + 175.16 driver (the last one).

On the other side MPC-hc + MPC decoder + nVidia = "seek" problem.

So, the "seek" problem is in the MPC decoder, correct ? (and not on MPC-hc or nVidia).

Cyberlink decoder also does not seems to suffer the max 4 Ref Frames issue that MPC decoder suffers.
On the other side, I see framerate going down (and becoming jerky) with Cyberlink decoder and I do not see this problem on MPC decoder (despite in some cases I do not see this framerate issue in Cyberlink so there might be some restrictions to encode for Cyberlink).

Too many variables to work on ... unless people fix all them like Metalmania did.

crlorentzen

23rd May 2008, 22:15

Here is my testing

Windows XP SP3, Nvidia Geforec 9600 GT 175.16 WHQL, MPC 1.1.0.2 (build 518).

DXVAChecker
NVIDIA GeForce 9600 GT
ModeMPEG2_C: DXVA1, NV12, 720x480 / 1280x720 / 1920x1080
ModeMPEG2_D: DXVA1, NV12, 720x480 / 1280x720 / 1920x1080
ModeMPEG2_IDCT: DXVA2, NV12, 720x480 / 1280x720 / 1920x1080
ModeVC1_IDCT: DXVA1/2, NV12, 720x480 / 1280x720 / 1920x1080
ModeWMV9_IDCT: DXVA1/2, NV12, 720x480 / 1280x720 / 1920x1080
ModeVC1_MoComp: DXVA1/2, NV12, 720x480 / 1280x720 / 1920x1080
ModeWMV9_MoComp: DXVA1/2, NV12, 720x480 / 1280x720 / 1920x1080
ModeVC1_PostProc: DXVA1/2, NV12, 720x480 / 1280x720 / 1920x1080
ModeWMV9_PostProc: DXVA1/2, NV12, 720x480 / 1280x720 / 1920x1080
1720AC81-9D1B-4F63-9A37-4A88483D0B87: DXVA1, NV12, 720x480 / 1280x720 / 1920x1080
ModeH264_VLD_NoFGT: DXVA1, NV12, 720x480 / 1280x720 / 1920x1080

pirates_sample.mkv 1280x534 - blank screen
shooter_sample.mkv 1280x528 - blank screen
sw_2_sample.mkv 1280x544 - blank screen
x3_sample.mkv 1280x536 - blank screen

Would the issue be the resolutions? Does ATI support more resolutions? Can someone provide a DXVACheck output for an ATI HD card?

_Rumbel__Soul_Eater_-_01__H264-HD_-001.mkv 1280x720 - works, no seeking
http://rapidshare.com/files/116651035/_Rumbel__Soul_Eater_-_01__H264-HD_-001.mkv

Apple Trailer(s) I have tested a few they look the same, here is an example
http://www.apple.com/trailers/weinstein/diaryofthedead/hd/
diaryofthedead-tlr1_h720p.hdmov 1280x720 - only top displays, seeking fails.
http://img87.imageshack.us/img87/2852/diaryofthedeaddxvatestbm5.th.jpg (http://img87.imageshack.us/my.php?image=diaryofthedeaddxvatestbm5.jpg)

Update: with Cyberlink's CL264dec.ax 2.1.0.1310 PowerDVD 8 the _sample.mkv files play and seek, however, the video gets "jittery"...the Soul Eater does not work excpet without DXVA, and the Apple Trailer test looks perfect, however, I think it has been said that Cyberlink may use DXVA1, so maybe not be an apples to apples comparison but, my CPU Utilization does sit at 0% with DXVA and 8% without using the Cyberlink decoder.

MarcioAB

23rd May 2008, 22:42

Would the issue be the resolutions?
No. For MPC decoder (MPCVideoDec.ax or internal into MPC) the black screen issue is related with the number of Ref Frames set during encode (it will work for encodes made with 4 or less Ref Frames).

For those 4 samples, the Ref Frames are 8, 8, 9, 9.
For the Rumbel, Ref Frames = 3.

You can quickly check the number of Ref Frames in an encode using MediaInfo before to play for no surprises ;-)

carnage_pl

24th May 2008, 18:57

_Rumbel__Soul_Eater_-_01__H264-HD_-001.mkv 1280x720 - works, no seeking

on ATI HD2300 dxva and seeking works fine.
but I have other problem.

General
Complete name : black.hawk.down.2001.1080p.bluray.x264.sample-wpi.mkv
Format : Matroska
File size : 52.0 MB
Duration : 1min.
Overal bit rate : 7198 Kb/s
Encoded date : UTC 2008-05-03 08:23:32
Writing application : mkvmerge v2.2.0 ('Turn It On Again') built on Mar 4 2008 12:58:26
Writing library : libebml v0.7.7 + libmatroska v0.8.1

Video
Format : AVC
Format/Info : Advanced Video Codec
Format profile : High@L5.1
Format settings, CABAC : Yes
Format settings, ReFrames : 5
Codec ID : V_MPEG4/ISO/AVC
Duration : 1min.
Bit rate : 6264 Kb/s
Width : 1920 piksele
Height : 800 piksele
Display aspect ratio : 2.400
Frame rate : 23.976 r/s
Colorimetry : 4:2:0
Scan type : Progressive
Title : Video

DXVA and seeking works but playing is slowed down compared to audio.
sample : black.hawk.down.2001.1080p.bluray.x264.sample_wpi.... (http://www.sendspace.pl/file/ZEecn12W/)

Mobility Radeon X2300 HD
ModeMPEG2_IDCT: DXVA2, NV12, 720x480
ModeMPEG2_A: DXVA1, NV12, 720x480
ModeMPEG2_C: DXVA1, NV12, 720x480
BB0796AE-2ED4-468D-A182-38F2CEADECF8: DXVA2, NV12, 720x480
ModeH264_MoComp_NoFGT_Avivo: DXVA2, NV12, 720x480
ModeH264_VLD_NoFGT_Avivo: DXVA2, -, -
ModeH264_VLD_NoFGT: DXVA2, NV12, 720x480 / 1280x720 / 1920x1080

Kado

24th May 2008, 23:41

No problems here with "soul eater" and "black hawk down" samples. Both play, seek and don't get out of sync with mpc internal dxva and cyberlink's decoder from pdvd8 using dxva.
vista sp1 + evr pre-sp1 + 9800gtx 175.63 + xxl's mpc rev 518.

carnage_pl

25th May 2008, 15:31

Kado

25th May 2008, 19:59

No slowdown here, I always use EVR custom presenter because the regular does not support subtitle rendering.

carnage_pl

25th May 2008, 20:58

maybe it's because old drivers. to mobile version of ati hd there aren't any new drivers.i tried 8.4 modded by dh mobility modder but it broke dxva. (windows crash). direct x updated to the newest. is it something what i can do?

ranpha

26th May 2008, 07:44

Testing with ATI Radeon HD 3200 (the 780G chipset) with Catalyst 8.5, Windows Vista Ultimate 64-bit and latest beta of MPC-HC using the internal decoder. I have a Phenom 9500, but I set the internal decoder to use only 2 threads to simulate a typical dual core setup.

http://www.awooga.org/images/9163kzpo3ud6jgbz8v.jpg (http://www.awooga.org/)

Test file one, the Haruhi OP. Tested with no subtitles because the internal subtitle engine takes more CPU time than DirectVobSub and may skew reading.
http://www.awooga.org/images/tbsn6zb4gvyi2ysubwn_thumb.jpg (http://www.awooga.org/viewer.php?file=tbsn6zb4gvyi2ysubwn.jpg)
13% CPU utilization average. If you ask me, DXVA works with this one.

Test file one, the Haruhi OP. Same case but DXVA disabled.
http://www.awooga.org/images/dn1vn9v7q43tyy4cgrt_thumb.jpg (http://www.awooga.org/viewer.php?file=dn1vn9v7q43tyy4cgrt.jpg)
Consistently above 20% CPU. In screenshot, it is 27%, more than double than when DXVA enabled.

Test file two, pirates_sample.mkv
http://www.awooga.org/images/8djrxxte5drmff77nri0_thumb.jpg (http://www.awooga.org/viewer.php?file=8djrxxte5drmff77nri0.jpg)
Played without any issues with DXVA working. CPU utilization is just like Haruhi OP.

Test file two, shooter_sample.mkv
http://www.awooga.org/images/ntri6tn0qwhj93oza16_thumb.jpg (http://www.awooga.org/viewer.php?file=ntri6tn0qwhj93oza16.jpg)
Also played with no issues with DXVA and with Haruhi's CPU utilization.

Test file three, sw_2_sample.mkv
http://www.awooga.org/images/aqr6r723n5716ixt4_thumb.jpg (http://www.awooga.org/viewer.php?file=aqr6r723n5716ixt4.jpg)
Also the same case as above.

Test file four, x3_sample.mkv
http://www.awooga.org/images/tp089q7245llttwric80_thumb.jpg (http://www.awooga.org/viewer.php?file=tp089q7245llttwric80.jpg)
The same case applies.

Will test with other files and append to this post later.

added:-

Test file five, Soul Eater clip with no subtitles.
http://www.awooga.org/images/bepw94haawhnas0wfz8r_thumb.jpg (http://www.awooga.org/viewer.php?file=bepw94haawhnas0wfz8r.jpg)
Played without any seeking troubles, with CPU utilization similar to Haruhi's with DXVA.

Test file six, the Black Hawk Down 1080p.
http://www.awooga.org/images/5f7ucj5inupp8su38v8y_thumb.jpg (http://www.awooga.org/viewer.php?file=5f7ucj5inupp8su38v8y.jpg)
Also played with no issues, with slightly higher CPU utilization (4 or 5 percent) than Haruhi despite being higher resolution. But did has some seeking trouble if I move the slider in the middle third of the clip, where audio goes out of sync before syncing itself a few second later. Moving the slider at the first third or the last third is problem-free.
The internal decoder still has a long way to go though for 1080p, which support is spotty at best, but for 720p files, MPC-HC has improved considerably from the last 3 months.