The May 2012 Open-Source Radeon Graphics Showdown

[curaga]

On my pc with radeon 9550 `mplayer -vo xv` works much faster than `mplayer -vo vdpau`, though my cpu is slow.

Why are you surprised? It's one extra layer compared to XV, it can never be faster until it accelerates the codec well enough.

[bug!]

On the newer oss drivers that support the vdpau state tracker, vdpauinfo reports that acceleration is in place and works.

On my pc with radeon 9550 `mplayer -vo xv` works much faster than `mplayer -vo vdpau`, though my cpu is slow.

[log0]

Only etqw-demo has problem, other games and test runs fine.

I build livecd with lastes mesa-git, xf86-drivers-git, kernel-git every 14-20 days for over a year (lastes build includes mesa-git 20120518), I tweaked mesa,kernel and drivers for maximum performance and do some tests, and have never seen so lightsmark showed less than 100 FPS, how Michael get 63 FPS, I do not know.
The-May-2012-Open-Source-Radeon-Graphics-Showdown radeon-tweaked does not show the the real data, certainly not for the HD 6770.

I'd test with latest mesa and report the segfault.

I'll do a run with your live cd. From here? http://www.gearsongallium.com/

[Pontostroy]

What about the other tests? Do they also crash?

Get a backtrace from the segfault and report to xorg/mesa guys.

I've got a 4850 available for testing here but never ran mesa from git.

Edit: Have you tried the latest git revision or the one Michael has used?

Only etqw-demo has problem, other games and test runs fine.

I build livecd with lastes mesa-git, xf86-drivers-git, kernel-git every 14-20 days for over a year (lastes build includes mesa-git 20120518), I tweaked mesa,kernel and drivers for maximum performance and do some tests, and have never seen so lightsmark showed less than 100 FPS, how Michael get 63 FPS, I do not know.
The-May-2012-Open-Source-Radeon-Graphics-Showdown radeon-tweaked does not show the the real data, certainly not for the HD 6770.

[log0]

I turn it on for last test (etqw-demo 1920x1080) with SwapBuffersWait off etqw-demo segfaults after 10-12 sec.

http://openbenchmarking.org/system/1...770/Xorg.0.log

What about the other tests? Do they also crash?

Get a backtrace from the segfault and report to xorg/mesa guys.

I've got a 4850 available for testing here but never ran mesa from git.

Edit: Have you tried the latest git revision or the one Michael has used?

[Pontostroy]

I did a quick diff of xorg.0.log.

It looks like you are running with SwapBuffers wait enabled and getting better results?

I turn it on for last test (etqw-demo 1920x1080) with SwapBuffersWait off etqw-demo segfaults after 10-12 sec.

http://openbenchmarking.org/system/1205192-BY-GOG20120574/Radeon%20HD6770/Xorg.0.log

[log0]

http://openbenchmarking.org/result/1...BY-GOG20120520

PTS run from my livecd, so anyone can reproduce the results.

I did a quick diff of xorg.0.log.

Michael:

(II) Module exa: vendor="X.Org Foundation"
	compiled for 1.11.3, module version = 2.5.0
	ABI class: X.Org Video Driver, version 11.0
(II) RADEON(0): KMS Color Tiling: enabled
(II) RADEON(0): KMS Pageflipping: enabled
(II) RADEON(0): SwapBuffers wait for vsync: disabled

Pontostroy:

(II) Module exa: vendor="X.Org Foundation"
	compiled for 1.12.1, module version = 2.5.0
	ABI class: X.Org Video Driver, version 12.0
(II) RADEON(0): KMS Color Tiling: enabled
(II) RADEON(0): KMS Pageflipping: enabled
(II) RADEON(0): SwapBuffers wait for vsync: enabled

It looks like you are running with SwapBuffers wait enabled and getting better results?

[Gusar]

xorg.conf isn't and has never been obsolete. You use xorg.conf.d now, but the idea is the same. You only need to include the Device section for the graphics driver. Everything else is autodetected (unless you want to customize those things too and then you can include those in xorg.conf.d).

This.

What's obsolete is the everything-including-the-kitchen-sink xorg.conf. Now you create just the section where you're configuring something. So if it's disabling SwapBuffersWait you want, you create the Device section and only that. Also, you now have the flexibility of multiple config files. If it makes it easier to wrap your head around it, think of xorg.conf as xorg.conf.d/99-something.conf

True, but i use much more slower cpu, pcie_gen2=1, vblank_mode=0, "SwapbuffersWait" "0" and have more FPS. Must be a reason.

Hmm, Unity vs. non-Unity?

[Pontostroy]

Can you test with openbenchmarking and upload link to your results please?

http://openbenchmarking.org/result/1205193-BY-GOG20120520

PTS run from my livecd, so anyone can reproduce the results.

[log0]

May be someone know: Why so big difference between propriatiry and opensource drivers? Something was not implemented or only not fully optimized...

The few developers are too busy writing the drivers to even think about optimization. 30%-60% speed is the maximum we can expect I guess.