Is AMD ever going to fix the critical video playback issue with Radeon HD2600 cards?

[tuxdriver]

Is AMD ever going to fix the video playback issues with Radeon HD2600 series of cards on both Windows and Linux?

This is what I'm talking about:

Here's a screenshot of a HD2600 card playing a video (Windows XP SP3) with the VMR renderer (the only way to get DXVA video acceleration under XP):



The video is totally screwed up and CPU usage goes sky high (80-100% CPU),way higher that with the legacy renderer.

On Linux,the problem with Xv playback in Xine is a bit different:
Symptoms are zoomed in video with thick "scanlines".
(sorry,I can't provide a Linux screenshot at the moment)

Looks like some kind of bug in the AMD drivers making these cards unusable for video playback (which is their main purpose)

[UPDATE]: The video corruption under WinXP is not a bug (it's fixed now).The only real issue is the broken WMV acceleration (high CPU usage)

[tuxdriver]

I know this is definitely a driver issue,because it worked when I was using Xorg 7.2,an old 2.6.21.7 kernel and Catalyst 8.4.

Xv video in Xine worked perfectly for about half an hour and then the output became corrupt.
Since then I've moved to a newer Xserver / kernel and tried every driver release,but the only thing I get is corrupted graphics with Xv.

The Windows XP DXVA problem is there since the card was released (I have tried every driver available,including all betas and hotfixes,but still no improvement)

[bridgman]

I haven't heard about video issues like this on Windows; have you talked to our cust support folks already ?

My initial impression was that VMR was kind of obsolete; I did a quick Google search and it seems to be borked on a lot of cards, not just ours. I'll try to talk to our MM folks tomorrow and see what the story is. EDIT - there also seem to be a bunch of VMR flavours, some of which work ("Renderless" ?) and others which do not. I don't really know enough about DXVA to give good answers right now, so stay tuned.

If you haven't reported this to our CS folks or checked the knowledgebase for Windows it would be a good idea. The general sense I got was that this all is kinda broken on Xp but generally works better on Vista.

I'm not aware of any hardware issues in the 6xx GPUs that would cause something like this.

[tuxdriver]

Yes,I'm using VMR9 (renderless) on XP.
It's the preferred mode for Windows Media Player and Media Player Classic.

VMR7 may be obsolete,but VMR9 is still popular/used by most players on XP.


bridgman,could you please report this to the Windows driver team as well:

http://www.vgaforum.com/showthread.php?t=21035

It's a nasty (but easy to fix) bug in the SSE detection code of the Windows driver that affects all AGP cards on CPUs that support SSE1 (and SSE2) instructions only (mainly AMD CPUs). More details and an unofficial fix are available in the thread.

[Spitfire]

Sorry, I don't understand how to reproduce this? I never had this issue with my hd2600xt. Can you help me in provoking this behavior to double check?

[tuxdriver]

Is your 2600XT an AGP or PCI-E version?

To experience the first (Windows) issue,just play a HD WMV wideo (not VC-1) in Windows Media Player on a fresh install of XP.

Here's a quick way to reproduce the video playback issue on Windows: Download the MotoGP '08 demo available at nVidia's nZone page: http://www.nzone.com/object/nzone_motogp08.html
Just install the game demo,go to the Movies folder and try to play any of those HD WMV videos with Windows Media Player or Media Player Classic)

For the Linux issue,I'll have to try Catalyst 9.1 again (this time with some kernel tweaks) and see if there's an improvement. Stay tuned.

[tuxdriver]

The bad thing about the newest Linux driver (9.1) is that I had to do a kernel recompile with this option:
'CONFIG_PCI_MSI=y'
to finally get this driver working.

It wouldn't run at all without this tweak (installs fine,but you get a black screen)

8.11 and 9.1 are the only drivers affected by this issue,but 8.12 works without the MSI switch.Weird.
I don't think any 'average' user would want to go through all this stuff.
I hope the next release wouldn't require this setting.

I still have to do a few more tests with various media players and desktop environments so keep an eye on this thread.

[PuckPoltergeist]

The bad thing about the newest Linux driver (9.1) is that I was forced to do a recompile with this option:
'CONFIG_PCI_MSI=y'
to finally get this driver working.

It wouldn't run at all without this tweak (installs fine,but you get a black screen)

8.11 and 9.1 are the only drivers affected by this issue,but 8.12 works.Weird.
I don't think any 'average' user would want to go through all this stuff.
I hope the next release wouldn't require this setting.

I still have to do a few more tests with various media players and desktop environments,so stay tuned.

8.12 depends also on MSI

[Spitfire]

Is your 2600XT an AGP or PCI-E version?

PCI-E, Sapphire HD2600XT 256MB DDR4

To experience the first (Windows) issue,just play a HD WMV wideo (not VC-1) in Windows Media Player on a fresh install of XPSP3.

Here's a quick way to reproduce the video playback issue on Windows: Download the MotoGP '08 demo available at nVidia's nZone page: http://www.nzone.com/object/nzone_motogp08.html
Just install the game demo,go to the Movies folder and try to play any of those HD WMV videos with Windows Media Player or Media Player Classic)

For the Linux issue,I'll have to try Catalyst 9.1 again (this time with some kernel tweaks) and see if there are any improvements. Stay tuned.

It's hard for me to reproduce this that way. I don't have SP3 and I cannot install that game at the moment. I have no disk space left, as I have a very small partition for Windows. Do you think this should also happen on SP2? And do you have a link to a small video of that type?

For Linux I can do better testing, if you found this happening there again.

[tuxdriver]

8.12 depends also on MSI

But strangely,8.12 *works* on a kernel compiled without this switch,unlike 8.11 and 9.1.