Tweaking KDE's KWin For Linux Gaming Performance

[Pontostroy]

My results
HD6770
pcie_gen2=1
swapbufferswait=0
Lightsmark 1600x900 window
Xrender - 142
opengl1 - 148
opengl2 - 145
opengles2 - 144
suspend - 175
openbox - 175

[bug77]

If testing OpenGL, why not test using a driver that properly supports it (e.g. nvidia blob)?
Any way, it looks that even with compositing disabled, KDE is still the slowest of the bunch when it comes to gaming.

As a side node, using Kubuntu 12.04 beta here and the option to suspend compositing in full screen was enabled by default.

[russofris]

I would also be curious (in your next WM comparison) to see a couple benchmark graphs with a "no windows manager" column. It's not that I believe that there are gamers that use Ctrl-Alt-F# as a WM, it's just that I would like an rational baseline for comparison, and gain the ability to assign a cost to each WM.

I guess that the kiosk, showcase, and arcade guys would be interested as well.

[Qaridarium]

after reading this article i chance my setting from openGL to xrender

[markos]

after reading this article i chance my setting from openGL to xrender

No need... The whole comparison is moot. Look again at the results, Gallium3D "Suspend effects" is capped around 60 fps. That's why xrender seems faster.

[darkbasic]

The real question is: why is it capped @60fps?

[kraftman]

Thank you very much for this interesting benchmarks and for listening to your readers Michael.

I did some benchmarks with Unigine Heaven and Catalyst and the results were identical in every possible scenario. When using Open Source Radeon drivers suspending compositions was helpful. Maybe there's something wrong with Intel?

[mrbumpy409]

The real question is: why is it capped @60fps?

Vsync? Perhaps the Vsync option is only kicking in when the Kwin desktop effects are suspended. I wonder if enabling Vsync in the KDE desktop effects settings would create a similar cap on the game's framerate with the desktop effects enabled.

[molecule-eye]

Articles like this are great. I wonder, however, if there are any differences between the compositing modes, i.e. between crisp, accurate, etc. I'm guessing that, if there are, they're pretty marginal.

The good news is that those of us using the proprietary Catalyst drivers don't have to worry about disabling compositing to get better performance.

[molecule-eye]

If testing OpenGL, why not test using a driver that properly supports it (e.g. nvidia blob)?

In exactly what sense do the Catalyst drivers not properly support OpenGL on linux? Just curious.