AMD to Demo Fusion APUs at IFA Berlin, in September

[allquixotic]

There are known GPU hangs when running 2D and 3D together - agd5f and Richard are both looking into those. AFAIK they also happen when using just 2D or just 3D but *much* less frequently. Probably would be a good idea to turn off compositing until the GPU hangs are figured out.

Not sure if the devs are seeing the same degree of corruption on their systems (at least I don't remember seeing or hearing about it); what are others seeing ?

I can echo Wingfeather's problems, except on a HD5970, using the latest linux-next kernel (alternately I tried airlied's drm-radeon-testing and drm-fixes) and evergreen_accel. OTOH I am running everything from master, including xserver, libs, and proto. Definitely seeing pointer corruption unless Option "RenderAccel" is set to "off", and random crashes whether compositing is enabled or not. I can also reliably crash X by minimizing a window with metacity or compiz compositing, although with Kwin compositing set to always keep thumbnails, that doesn't happen. Tried a full factorial of the radeon DDX's options, and only RenderAccel off improved things at all. "EXAVSync" "on" causes X to crash on startup.

[bridgman]

OK, that seems like a useful clue. Do you still get hangs eventually when running the open drivers after fglrx ?

Which GPU are you using ?

EDIT - I guess we should move this to the HD5000 open source driver thread for now

[monraaf]

Evergreen problems

I was planning to do a detailed writeup, but I see others have beaten me to it One other thing that I'm noticing is that if I cold boot I see an almost immediate GPU lockup at the GDM login screen. It seems that I have to run fglrx first to set the GPU to a sane state before I can use any of the open source acceleration stuff.

[Wingfeather]

Not much to add, but I've tried it today with xorg-server 1.8.2 and it seems to crash more than with 1.7.7. Unless I add the "RenderAccel" "off" option I can't even get X to start up with either KDE or E.

Once in, doing anything 3D at all (glxgears or turning on compositing) results in an immediate hang of X every time. Cursor still moves, but can't light the capslock or do anything.

[rohcQaH]

What problems are you seeing ?

I've written about my experience here (all of those were confirmed by others in the thread) and updated it today here.

the good news is that I had zero system freezes or X crashes today, but it's still far away from a working system.


I did ask which of those I should turn into detailed bug reports, but received no answer. If you claim not to know about those problems, the answer would be "all of them", right?

[bridgman]

There are known GPU hangs when running 2D and 3D together - agd5f and Richard are both looking into those. AFAIK they also happen when using just 2D or just 3D but *much* less frequently. Probably would be a good idea to turn off compositing until the GPU hangs are figured out.

Not sure if the devs are seeing the same degree of corruption on their systems (at least I don't remember seeing or hearing about it); what are others seeing ?

[Wingfeather]

I've tried it on my 5850 (both the evergreen_accel branch of radeon and the latest mesa), and X (1.7.7-r1) does run with it. I get all right noises in the xorg log file and glxinfo, too, which is encouraging.

However, in terms of actual performance, I get:
- Corrupted cursor, always
- Full-screen corruption of essentially all desktop elements, in a way which change as I things move around. Really bad corruption, this is. Text is effectively unreadable wherever it is on the screen.
- Random crashes of both X and entire system
- Didn't try videos

I tried enlightenment E17 and KDE-4.4.5, both with openGL compositing and XRender. One thing I can say is that while corruption is severe with both, it is much worse with XRender.

A random thing is that once I'd run X with hardware acceleration and had all the corruption, switching back to software rendering didn't get rid of it. It needed a reboot to sort it out.

Is this what other people are getting?

[bridgman]

Hmm okay. Then who is working on getting the open source driver stack working with the Evergreen parts? Because right now it's not really working.

What problems are you seeing ?

[monraaf]

Alex (agd5f) is working on getting the open source driver stack working with the Bobcat-based part (Ontario) now.

Hmm okay. Then who is working on getting the open source driver stack working with the Evergreen parts? Because right now it's not really working.

[bridgman]

Just a minor point - the article suggests that this is the first Fusion demo ("will finally make a demo"), but the first two Fusion parts (Ontario and Llano) were already demonstrated at Computex back in early June :

http://www.semiaccurate.com/2010/06/02/amd-fusion-demo-video/

Alex (agd5f) is working on getting the open source driver stack working with the Bobcat-based part (Ontario) now.