Log in

View Full Version : New DXVA/GPU accel. discussion (from CoreAVC)

Pages : 1 2 [3]
Jay Bee
26th May 2007, 10:16
then decoder fails to connect to video renderer on dxva mediatype. it hardware acceleration option applied? I mean when you switch it to 'on' mode and push the 'apply' button it was still 'on'?

Yes, I rechecked the option during playback and it was still on.
KoD
26th May 2007, 11:09
On the level of hardware assisted AVC decoding; Dont set your hopes on Ati's UVD yet. Seems they made wrongfull claims about the 2900XT having the UVD-engine on-board.

Quote from the article;


The 2900XT is the only new-gen Ati board available atm. The others will come in the following months. So anyone that has bought himself one of the 2900XT's, and hopes to use that board for AVC decoding, might (for now) be out of luck ....

Read te full article on dailytech;
http://www.dailytech.com/article.aspx?newsid=7447

Only the 2600 and 2400 have a dedicated video chip (the mobile versions will have it as well). For the 2900 everything is done using the shader units, no additional chip will be used. Take everything that websites say with a grain of salt.
Theliel
26th May 2007, 13:03
the same here, same test and YUY2.
Ronin-7
26th May 2007, 13:14
The 2900 definitely doesn't have UVD instead it has the same level of GPU acceleration as the previous X1K line (now branded as Avivo HD).

Unfortunately for AMD they really dropped the ball on their marketing materials as the fact that the 2900 did not support UVD was tucked away on another slide in the HD 2900 series presentation whereas other slides gave the impression it was present on all 2xxx series GPU's & to make matters worse some 2900XT's shipped with the wrong packaging advertising UVD support before the corrected labels where shipped to their partners.

AMD's own tech rep on Beyond3D & the Tech Report have confirmed that the 2900XT does not support UVD.

As for anyone wondering why well transistor budgets and the fact that the majority of people who would end up buying a 2900XT would have a fairly beefy CPU to begin so UVD is not as badly needed. Pretty much the same deal with the 8800GTX and Purevideo VP2 only on the 8600/8500 lines though NVIDIA didn't shoot themselves in the foot unlike AMD when it came to marketing.
mariner
26th May 2007, 16:28
Hi Sergey,

Greetings and thank you for gracing the forum again.

Tested the MainConcept H.264 Decoder v3.0.7299 and Cyberlink H264 Decoder v2.0.0.1709 with 6600GT/93.71.
The MainConcept decoder only managed half the frame rate of Cyberlink enen with deblocking turned off. How can the performance be improved?

Thanks and best regards.
Jay Bee
26th May 2007, 16:29
I see GPU manufacturers are living up to their reputation of lying to their customers and internet "tech" websites are living up to their reputation of not having a clue about anything but gaming.
Crisidelm
26th May 2007, 19:33
Elecard DXVA on (http://www.crisidelm.it/images/ElecardAVC_1.gif)
Output format with DXVA on (http://www.crisidelm.it/images/ElecardAVC_2.gif)
Elecard DXVA off (http://www.crisidelm.it/images/ElecardAVC_3.gif)
Output format with DXVA off (http://www.crisidelm.it/images/ElecardAVC_4.gif)
Sergey A. Sablin
27th May 2007, 06:22
Elecard DXVA on (http://www.crisidelm.it/images/ElecardAVC_1.gif)
Output format with DXVA on (http://www.crisidelm.it/images/ElecardAVC_2.gif)
Elecard DXVA off (http://www.crisidelm.it/images/ElecardAVC_3.gif)
Output format with DXVA off (http://www.crisidelm.it/images/ElecardAVC_4.gif)

1. try to use latest drivers
2. try to disable deblock (set it to Off) and compare performance in software mode and in DXVA mode
KoD
27th May 2007, 10:16
Off topic:

I see GPU manufacturers are living up to their reputation of lying to their customers

It's not lying, it's called "marketing".

Like shampoo ads that promise "your hair will be 60% stronger and it will shine twice as more".

But obviously nobody is going to test the fake claims of a shampoo ad.

However, when you get a video card and use a bechmark suite and see the real results are not what they advertised, one gets angry.

Lesson to be learned: the marketing strategy for products that consumers take pride in testing the claims and show off on the internet should be a little different than for products consumers don't.

Disclaimer: please don't throw rotten tomatoes at me ! ^^


P.S. : well, to be honest I think I remember in the UK there are laws that allow one to sue a company for such fake claims and I believe at least L'Oreal was convicted once for a shampoo ad. They did the mistake to say "scientific experiments proved that L'Oreal xxx shampoo makes your hair yyy% stronger .. and I don't know what else". But when it came the moment to show in court evidence of such tests being performed by an independent testing facility, they had nothing to show as they did none. Still, many people don't go that far and many ads don't make the mistake of claiming "scientific experiments have been done".
CruNcher
27th May 2007, 12:29
Off topic:
It's not lying, it's called "marketing".


Yes and No the definition of marketing isn't lying at all but it more and more becomes that, someone that studies marketing will tell you the same it's more a tool to show the consumer what a product can do and what not and especialy to advertise features that other products don't have on the market, but when you missuse marketing in my eyes it's lying to the consumer and especialy lying to yourself as a marketing guy, but hey that's the american way of life lying to yourself every day ;-)
Crisidelm
27th May 2007, 16:40
By the way, I can see that with the latest Mainconcept H.264 codec when switching DXVA on the output format is NV12.
Theliel
27th May 2007, 17:45
version?
Crisidelm
27th May 2007, 18:35
3.0.7299
When installing the aforementioned elecard H.264 codec, the Mainconcept H.264 codec disappears (as if it was automatically deregistered from the OS). When I re-register the Mainconcept H.264 codec, the elecard gets automatically unregistered. But it looks like the DXVA works (with the same problems of deblocking that you can experience with the Elecard as Sablin said already) now.
pankov
29th May 2007, 23:12
Guys, does anybody know if I can profit from the hardware acceleration of the 8500/8600 series in the following scenario
1. one MSI 8600GTS (fanless) in my primary PEG slot (16xPCI-E)
2. one (NV 7xxx) in my secondary PEG slot (16xPCI-E slot with 4x interface on P965 mobo)
3. TV/PJ connected to the 7xxx card so I can use PowerStrip
4. and of course PowerDVD 7.3.xxxx on Vista 32-bit
Does video acceleration work only when playing the video stream on the 8500/8600 card or it's there for the decoder to use it despite the fact that it's not using the same card for output.
addit
2nd June 2007, 14:35
Pankov, i'm in the process of building a HTPC and i'm wondering the same thing. It would make life so much easier outputing through HDMI on one card and having a 8600 sitting there to accelerate the video :cool: Does anyone wether this is possible?
hdboy
7th June 2007, 05:43
i would think it'd be easier just to get a 8600 card with hdmi
Xin-Hong
7th June 2007, 16:53
Hi Sergey,

Greetings and thank you for gracing the forum again.

Tested the MainConcept H.264 Decoder v3.0.7299 and Cyberlink H264 Decoder v2.0.0.1709 with 6600GT/93.71.
The MainConcept decoder only managed half the frame rate of Cyberlink enen with deblocking turned off. How can the performance be improved?

Thanks and best regards.

I have a 6600GT too under Vista, does the 6600GT with dxva reduce the %CPU in H.264 movies ? and how much ?

Thanks :)

Edit : I tried with some HD Trailers and I am always in YUY2 so the 6600GT does'nt do anything, i tried in VMR and EVR.

http://pix.nofrag.com/d8/70/1cefdc487e16128b2096d38e2cfd.jpg
NanoBot
8th June 2007, 01:21
Hi,

I would like to report that full h.264 decoding in the GPU with NVidia 8500/8600 based graphics cards is already possible under Windows XP, but it is still beta. You need

Cyberlinks newest h.264 decoder filter ( cl264dec.ax ) version 2.0.0.1709, which is part of PowerDVD 7.3 Ultra, build 2911

and

NVidias beta driver "Forceware 165.01" for Windows XP. Since this driver is not listed on the official site, you have to google for it, try "165.01 XP" and you will find it.

I tested this codec/driver combination on an Asus 8500GT card with DVBViewerPro, AltDVB 2.2 und Casimirs media player classic mod and got CPU loads between 8% and 25% on an AMD64 x2 3800+. When using the overlay renderer the peak CPU load of course was a little bit lower then VMR9. But even with VMR9 and subtitles activated, I never saw CPU loads above 25% peak.

But remember, the driver is beta and is far away from working without problems. So do not be surprised to see freezes during playback or a completly freezed system from time to time :-)

C.U. NanoBot

P.S.: The video material tested is h.264 video with a resolution of 1920 x 1080i, the bitrate is 17MBit/s - 20MBit/s.
Sirber
8th June 2007, 01:57
On card and CPU that you have, if it was GPU accelerated CPU load would be bellow 5%.
DXVA2 don't work in overlay or in VRM9, only work in EVR.
NanoBot
8th June 2007, 03:51
Hi Sirber,

I forgot to say that the CPU load measured with DVBViewerPro and AltDVB includes the load of a special plugin ;-)
And yes, it is not DXVA2 what is used here, since this is not available under Windows XP. Cyberlink uses "internal" API-Calls from NVIDIA to make hardware decoding work under XP. Therefore under XP it works with both the overlay renderer and VMR9.

When playing back already decoded transport streams the CPU load is lower than stated above. What I found is that the CPU load using vmr9 is significantly higher than with overlay.

Does anybody know a program which is able to do a screenshot even when using overlay ? Then I could make a screenshot where you can see the CPU load during playback by yourself.

C.U. NanoBot
KoD
8th June 2007, 07:49
"CPU load using vmr9 is significantly higher than with overlay"

Amazing... you only found that when playing transport streams. Did it ever occure to you VMR9 is always slower than Overlay, no matter what video file you play ?
NanoBot
8th June 2007, 08:53
Hi KoD,

of course I know that the vmr9 renderer is eating more CPU power than the overlay renderer. My other graphic card is a 7900GT, and there I was used to see CPU loads about 50% - 70 % with and 70% - 90% without GPU support. In relation to the CPU load caused by the h.264 decoding process itself, the difference between vmr9 and overlay seems not to be so significant.

But now, with the much lower CPU loads due to full GPU support, I notice that difference much more than before.

C.U. NanoBot
mariner
8th June 2007, 15:39
Tested the H264 decoder from nero v 4.10.5.0 with DXVA turned on and no post processing. Frame rate achieved is again only half of Cyberlink's decoder.

Below is the list of subtypes supported by the three decoders. With my 6600GT/93.71 running under XP, the subtype used is always {70174349-D888-408A-BB1D-9E18075E5A5A}. Does anyone know the difference among the various DXVA/NV12 subtypes?

Thanks and best regards.


nero v4.10.5.0
Major Type: Video - Sub Type: {70174349-D888-408A-BB1D-9E18075E5A5A} - Format: DXVA 1920x1088, 0 bits,
Aspect Ratio: 1920x1088,
Interlace format: Interleaved Bob Only
rcSrc=(0,0,1920,1088)
rcDst=(0,0,1920,1088)
CtrlFlags=00000000
Major Type: Video - Sub Type: {839DC727-010F-441D-957C-A3C1CB15D950} - Format: DXVA 1920x1088, 0 bits,
Aspect Ratio: 1920x1088,
Interlace format: Interleaved Bob Only
rcSrc=(0,0,1920,1088)
rcDst=(0,0,1920,1088)
CtrlFlags=00000000
Major Type: Video - Sub Type: {5CD11EE4-DAC9-4449-A017-440B36C0DE3D} - Format: DXVA 1920x1088, 0 bits,
Aspect Ratio: 1920x1088,
Interlace format: Interleaved Bob Only
rcSrc=(0,0,1920,1088)
rcDst=(0,0,1920,1088)
CtrlFlags=00000000
Major Type: Video - Sub Type: {5CD11EE1-DAC9-4449-A017-440B36C0DE3D} - Format: DXVA 1920x1088, 0 bits,
Aspect Ratio: 1920x1088,
Interlace format: Interleaved Bob Only
rcSrc=(0,0,1920,1088)
rcDst=(0,0,1920,1088)
CtrlFlags=00000000
Major Type: Video - Sub Type: YV12 - Format: YV12 1920x1088, 12 bits,
Aspect Ratio: 1920x1088,
Interlace format: Interleaved Bob Only
rcSrc=(0,0,1920,1088)
rcDst=(0,0,1920,1088)
CtrlFlags=00000000
Major Type: Video - Sub Type: YUY2 - Format: YUY2 1920x1088, 16 bits,
Aspect Ratio: 1920x1088,
Interlace format: Interleaved Bob Only
rcSrc=(0,0,1920,1088)
rcDst=(0,0,1920,1088)
CtrlFlags=00000000
Major Type: Video - Sub Type: RGB32 - Format: RGB 1920x1088, 32 bits,
Aspect Ratio: 1920x1088,
Interlace format: Interleaved Bob Only
rcSrc=(0,0,1920,1088)
rcDst=(0,0,1920,1088)
CtrlFlags=00000000

mainconcept v3.0.7299
Major Type: Video - Sub Type: {70174349-D888-408A-BB1D-9E18075E5A5A} - Format: NV12 1920x1088, 12 bits,
Aspect Ratio: 4x3,
Interlace format: Weave Only
rcSrc=(0,0,1920,1088)
rcDst=(0,0,1920,1088)
CtrlFlags=00000000
Major Type: Video - Sub Type: {5CD11EE4-DAC9-4449-A017-440B36C0DE3D} - Format: NV12 1920x1088, 12 bits,
Aspect Ratio: 4x3,
Interlace format: Weave Only
rcSrc=(0,0,1920,1088)
rcDst=(0,0,1920,1088)
CtrlFlags=00000000
Major Type: Video - Sub Type: YUY2 - Format: YUY2 1920x1088, 16 bits,
Aspect Ratio: 4x3,
Interlace format: Weave Only
rcSrc=(0,0,1920,1088)
rcDst=(0,0,1920,1088)
CtrlFlags=00000000
Major Type: Video - Sub Type: UYVY - Format: UYVY 1920x1088, 16 bits,
Aspect Ratio: 4x3,
Interlace format: Weave Only
rcSrc=(0,0,1920,1088)
rcDst=(0,0,1920,1088)
CtrlFlags=00000000
Major Type: Video - Sub Type: RGB32 - Format: RGB 1920x1088, 32 bits,
Aspect Ratio: 4x3,
Interlace format: Weave Only
rcSrc=(0,0,1920,1088)
rcDst=(0,0,1920,1088)
CtrlFlags=00000000
Major Type: Video - Sub Type: RGB24 - Format: RGB 1920x1088, 24 bits,
Aspect Ratio: 4x3,
Interlace format: Weave Only
rcSrc=(0,0,1920,1088)
rcDst=(0,0,1920,1088)
CtrlFlags=00000000
Major Type: Video - Sub Type: RGB565 - Format: RGB 1920x1088, 16 bits,
Aspect Ratio: 4x3,
Interlace format: Weave Only
rcSrc=(0,0,1920,1088)
rcDst=(0,0,1920,1088)
CtrlFlags=00000000
Major Type: Video - Sub Type: RGB555 - Format: RGB 1920x1088, 16 bits,
Aspect Ratio: 4x3,
Interlace format: Weave Only
rcSrc=(0,0,1920,1088)
rcDst=(0,0,1920,1088)
CtrlFlags=00000000
Major Type: Video - Sub Type: YUY2 - Format: YUY2 1920x1088, 16 bits
rcSrc=(0,0,1920,1088)
rcDst=(0,0,1920,1088)
Major Type: Video - Sub Type: UYVY - Format: UYVY 1920x1088, 16 bits
rcSrc=(0,0,1920,1088)
rcDst=(0,0,1920,1088)
Major Type: Video - Sub Type: RGB32 - Format: RGB 1920x1088, 32 bits
rcSrc=(0,0,1920,1088)
rcDst=(0,0,1920,1088)
Major Type: Video - Sub Type: RGB24 - Format: RGB 1920x1088, 24 bits
rcSrc=(0,0,1920,1088

cyberlink v2.0.0.1709
Major Type: Video - Sub Type: {1B81BE68-A0C7-11D3-B984-00C04F2E73C5} - Format: dxva 1920x1088, 12 bits,
Aspect Ratio: 4x3,
Interlace format: Interleaved Bob Only
rcSrc=(0,0,0,0)
rcDst=(0,0,1920,1088)
CtrlFlags=00000000
Major Type: Video - Sub Type: {1B81BE69-A0C7-11D3-B984-00C04F2E73C5} - Format: dxva 1920x1088, 12 bits,
Aspect Ratio: 4x3,
Interlace format: Interleaved Bob Only
rcSrc=(0,0,0,0)
rcDst=(0,0,1920,1088)
CtrlFlags=00000000
Major Type: Video - Sub Type: {70174349-D888-408A-BB1D-9E18075E5A5A} - Format: DXVA 1920x1088, 12 bits,
Aspect Ratio: 4x3,
Interlace format: Interleaved Bob Only
rcSrc=(0,0,0,0)
rcDst=(0,0,1920,1088)
CtrlFlags=00000000
Major Type: Video - Sub Type: NV12 - Format: DXVA 1920x1088, 12 bits,
Aspect Ratio: 4x3,
Interlace format: Interleaved Bob Only
rcSrc=(0,0,0,0)
rcDst=(0,0,1920,1088)
CtrlFlags=00000000
Major Type: Video - Sub Type: {5CD11EE4-DAC9-4449-A017-440B36C0DE3D} - Format: DxVA 1920x1088, 12 bits,
Aspect Ratio: 4x3,
Interlace format: Interleaved Bob Only
rcSrc=(0,0,0,0)
rcDst=(0,0,1920,1088)
CtrlFlags=00000000
Major Type: Video - Sub Type: {BF22AD00-03EA-4690-8077-473346209B7E} - Format: DXvA 1920x1088, 12 bits,
Aspect Ratio: 4x3,
Interlace format: Interleaved Bob Only
rcSrc=(0,0,0,0)
rcDst=(0,0,1920,1088)
CtrlFlags=00000000
Major Type: Video - Sub Type: YUY2 - Format: YUY2 1920x1088, 16 bits,
Aspect Ratio: 4x3,
Interlace format: Interleaved Bob Only
rcSrc=(0,0,0,0)
rcDst=(0,0,1920,1088)
CtrlFlags=00000000
Major Type: Video - Sub Type: YV12 - Format: YV12 1920x1088, 12 bits,
Aspect Ratio: 4x3,
Interlace format: Interleaved Bob Only
rcSrc=(0,0,0,0)
rcDst=(0,0,1920,1088)
CtrlFlags=00000000
Major Type: Video - Sub Type: YVYU - Format: YVYU 1920x1088, 16 bits,
Aspect Ratio: 4x3,
Interlace format: Interleaved Bob Only
rcSrc=(0,0,0,0)
rcDst=(0,0,1920,1088)
CtrlFlags=00000000
Major Type: Video - Sub Type: RGB24 - Format: RGB 1920x1088, 24 bits,
Aspect Ratio: 4x3,
Interlace format: Interleaved Bob O
Sirber
8th June 2007, 15:44
Four 4:2:0 12-bpp formats are recommended, with the following FOURCC codes:

* IMC2
* IMC4
* YV12
* NV12

In all of these formats, the chroma channels are subsampled by a factor of two in both the horizontal and vertical dimensions.

http://msdn2.microsoft.com/en-us/library/ms867704.aspx
KoD
4th July 2007, 10:01
As usual, Beyond3D is the place for from the source info on graphic cards. Here is what they say about the Radeon 2400XT and 2600XT and video decoding acceleration (among other things): article (http://www.beyond3d.com/content/reviews/28/1).

In another article, it is said that the 2900 should support all that the UVD does using its shader units, but this will have to wait for driver support.
mariner
16th July 2007, 16:59
http://msdn2.microsoft.com/en-us/library/ms867704.aspx

Hi Sirber, thanks for the link.

I was not able to find anything on deciphering the following DXVA subtypes. Any help is most appreciated.

{1B81BE68-A0C7-11D3-B984-00C04F2E73C5}
{1B81BE69-A0C7-11D3-B984-00C04F2E73C5}
{70174349-D888-408A-BB1D-9E18075E5A5A}
{5CD11EE4-DAC9-4449-A017-440B36C0DE3D}
{BF22AD00-03EA-4690-8077-473346209B7E}
{1B81BEA0-A0C7-11D3-B984-00C04F2E73C5}
{1B81BEA1-A0C7-11D3-B984-00C04F2E73C5}
{1B81BEA2-A0C7-11D3-B984-00C04F2E73C5}

Best regards.
bnz
11th October 2007, 12:20
Having a new HD2600pro card running with Vista, can anybody give me an explanation why dxva hardware acceleration (using the cyberlink decoder) does not work with rendering modes such as VMR9 when it seems to work with XP? It doesn't make sense to me...
doctorcilantro
24th December 2007, 05:23
Same here! Just upped from an AGP 2600xt and XP with DXVA working, and now my 3850 PCI-E system with Vista x64 won't accelerate the same files the XP system could....wtf?

DC
arfster
24th December 2007, 10:48
VMR9 acceleration works fine for mpeg2/vc1, but you need to use EVR for h264 (Vista doesn't support dxva1>dxva2 translation for it).

Of course, AGP acceleration is extremely troublesome so your problem could be drivers, and Vista64 creates even more issues. Frankly I'd be impressed if you get that combo to work.
doctorcilantro
26th December 2007, 03:44
I need to try 7.7 on Vista x64 and see if it works.

I switched to XP and the 3750 (PCI-E) now does DXVA on the same files the 2600xt (AGP) would.

DC