ATI R600/700 OSS 3D Driver Reaches Gears Milestone

[monraaf]

For me I haven't had the time to really test OpenGL games, but I have been running Compiz over a week now with this driver and it works great and unlike fglrx no crashes at all.

Now all AMD has to do is make hardware accelerated H.264 decoding available and I'm a satisfied customer

[pingufunkybeat]

For me, they are surprisingly stable. This is all still preliminary, experimental, and unreleased stuff, but it rarely crashes for me.

2D is rock solid. 3D hangs on a couple of things for me (KWin with OpenGL effects and foobillard), and a couple of things don't work (supertuxkart). But the things that work, work with no problems.

The major remaining bug is the graphical corruption in games like OpenArena. When using direct rendering, it is unplayable, along with some other geometrically more complex games. Still, it doesn't crash, at least for me.

The workaround is to run these games indirectly. Then they render perfectly, but do crash the X server from time to time. OpenArena crashes (kills X, basically) every time you win a level, for example. It works fine until then.

So I guess that there are two major bugs here, the corruption in the direct rendering mode, and the X crashes in the indirect mode. When they are finished, these things will be usable for daily work, and it will be bug-hunting time. The devs have been hunting the corruption one for the last few days, and there are first commits which considerably improve things. I think the bug is slightly better understood now, and with some luck, the devs will figure out how to fix it soon.

I don't know about the comparison to catalyst, as I've never run it. I imagine it's definitely inferior at this time, and it doesn't offer all the features (OpenGL 2+ is only coming with the KMS/GEM/DRI2/Gallium stack, for example).

EDIT: This is my understanding of the situation as an interested observer who runs the latest drivers from git and reads the IRC channel. I'm sure that the devs will be able to explain things in more detail.

[monkeynut]

So these drivers can do OpenArena now? Nifty...

A few questions:
How stable are they? (If unstable when do they crash, what are you doing when it occurs?)
What is performance like compared to catalyst?

Thanks for your time.

[Fran]

A workaround for the triangle mess is to start the game with "LIBGL_ALWAYS_INDIRECT=1 openarena". It reportedly also fixes other games, like vdrift.

Yeah, thanks, that was it. Quake3 1.34-rc3 works great now (leaving aside the random X restarts, of course ).

[pingufunkybeat]

I have only tried OpenArena, which is ioquake3.

A workaround for the triangle mess is to start the game with "LIBGL_ALWAYS_INDIRECT=1 openarena". It reportedly also fixes other games, like vdrift.

Unfortunately, it crashes from time to time when using LIBGL_ALWAYS_INDIRECT. It didn't use to crash in the direct mode (but then it's unplayable).

[Fran]

Great work, this is an exciting time! My first game of Quake3 on free drivers!

Which version of quake3 have you tried? In my case, 1.32c (binary) restarts X randomly when approaching a portal (and alsa doesn't work), otherwise it works great. With 1.34-rc3 (compiled) I get alsa sound, but it is unplayable (triangles everywhere!).

[pingufunkybeat]

The recent drm patch fixed the leaking, and neverball, supertux, etc. are perfect now.

OpenArena is still messed up (triangles all over the place), but if you run it with LIBGL_ALWAYS_INDIRECT=1, it actually runs perfectly (though it brought down X at the end of the level). I get playable framerates at 1920x1080 on my 4550, which is amazing.

I guess that there is a difference in how things are handled in direct and indirect mode. I assume that handling things directly (once the triangle issue is solved) will bring a slight speed improvement too.

Great work, this is an exciting time! My first game of Quake3 on free drivers!

[Fran]

And BTW (sorry for the spamming) I just updated to the latest drm/mesa/driver and Neverball is almost 100% glitch-free. Just a minor transparency issue with doors.

Also, the black-band flickering I got in Darwinia and Frets on Fire was caused by metacity compositing. With compositing disabled Darwinia's intro and world selection work quite well, but entering a world restarts X with

Backtrace:
0: /usr/X11R6/bin/X(xorg_backtrace+0x26) [0x4ec126]
1: /usr/X11R6/bin/X(xf86SigHandler+0x6f) [0x49a0cf]
2: /lib/libc.so.6 [0x7f9a875d6110]
3: /usr/lib64/xorg/modules/extensions//libglx.so [0x7f9a86ab73d4]
4: /usr/lib64/dri/r600_dri.so(__driUtilUpdateDrawableInfo+0xc1) [0x7f9a73260a71]
5: /usr/lib64/dri/r600_dri.so(radeonGetLock+0x68) [0x7f9a732829c8]
6: /usr/lib64/dri/r600_dri.so(rcommonFlushCmdBuf+0x21) [0x7f9a7327fc81]
7: /usr/lib64/dri/r600_dri.so(radeonDestroyContext+0x156) [0x7f9a7327f026]
8: /usr/lib64/dri/r600_dri.so [0x7f9a732608a0]
9: /usr/lib64/xorg/modules/extensions//libglx.so [0x7f9a86ab8cc9]
10: /usr/lib64/xorg/modules/extensions//libglx.so(__glXFreeContext+0x6c) [0x7f9a86aaf44c]
11: /usr/lib64/xorg/modules/extensions//libglx.so [0x7f9a86aaf483]
12: /usr/X11R6/bin/X(FreeClientResources+0xd3) [0x433f93]
13: /usr/X11R6/bin/X(CloseDownClient+0x60) [0x4463d0]
14: /usr/X11R6/bin/X(Dispatch+0x210) [0x44b4e0]
15: /usr/X11R6/bin/X(main+0x3b5) [0x431be5]
16: /lib/libc.so.6(__libc_start_main+0xfd) [0x7f9a875c2a3d]
17: /usr/X11R6/bin/X [0x431089]

Frets on Fire menu works, but starting a song hangs the game.

[Fran]

Ok devs, I don't need the manual fan control anymore, so you can continue improving the 3D part of the driver :P. If anyone else with a PCIe 3850 256MB BBA card wants the latest BIOS:

http://forums.techpowerup.com/showthread.php?t=58485

It regulates the fan speed while keeping it as quiet as possible. Be careful with the flashing


PD: Shit, I could have fixed my card 1 year and a half ago. But since I always had a means of regulating the fan speed manually (until now) I didn't do it. Sometimes the lack of features is good to search for better solutions :P

[Fran]

Whoa. I don't think one should have to do something like this. The card is supposed to regulate fan speed according to the heat/temp data that is coming from the sensor. If this is not hardwired then maybe some data is misinterpreted.

Probably not the problem in this case (1 min is too fast to overheat), but I'm positive my card doesn't regulate the fan speed. Neither in windows or in linux. In windows I use rivatuner to set the fan speed at 50-60% before starting a game, in linux I use aticonfig. If I don't do it, the game hangs (the temperature reaches >90?, and the fan stays at 16%). I think the first batch of Sapphire HD 3850 cards (reference cards, build by ati) have this problem. I don't know if they relased a fixed bios... I'll take a look.

(edit) Now that I think about it, maybe it did regulate the fan speed, but only if the GPU reached 100?C, and then it would just set the fan speed at 100% (that's like having a vacuum cleaner inside the case). Not a very smart regulation :P

(edit2): Seems like there ARE fixed bioses for this problem:

http://www.rage3d.com/board/showthread.php?t=33919389

Now I'll have to find one for my card (I assume any bios for a BBA card is ok, but better be sure).