The Cost Of ATI Kernel Mode-Setting On Fedora 12

We started by looking at the CPU usage of this triple-core AMD processor when playing back a 1080p H.264 video file, similar to our Intel Fedora 12 tests. However, first we looked when using the GL2 video output rather than X-Video. When using GL2 with mplayer, the lowest average CPU usage was actually with kernel mode-setting on Fedora 12, but Fedora 12 (UMS) and Fedora 10 had their CPU usage extremely close. Looking at the line graph, the lines for Fedora 12 (UMS) and Fedora 10 almost match up, but the Fedora 12 (KMS) run resulted in quite a radically different pattern.

Our second video-cpu-usage test profile run was with using the X-Video adapter where the CPU usage between all three runs were quite close and at about half the CPU usage rate as using GL2. The CPU usage lines between user and kernel mode-setting for Fedora 12 almost matched up precisely, while Fedora 10 had a number of CPU spikes.

Switching over to OpenGL gaming with Fedora 12, we first ran Urban Terror. Confirming what many people had reported in our forums and in bug reports, using the kernel mode-setting driver in Fedora 12 had resulted in a performance lost. In fact, it was quite a performance loss. When reverting to user-space mode-setting with the nomodeset option, Urban Terror at 800 x 600 had resulted in an average FPS of 76 FPS but 60 FPS when using the default KMS option. Fedora 10 was at 73 FPS even though it uses kernel mode-setting by default. When reaching 1920 x 1080 with Urban Terror, the Fedora 12 UMS was running at 73 FPS while the Fedora 12 KMS run had dropped even further to 46 FPS. For all resolutions tested (800 x 600, 1024 x 768, 1280 x 960, 1280 x 1024, 1400 x 1050, and 1920 x 1080), Fedora 12 UMS provided the best results followed by Fedora 10 and then Fedora 12 with KMS, when using the ATI Radeon X1800XT.