Description of problem: With kernel mode setting enabled (default) 2D performace is visibly lower than with nomodeset. The difference is clearly visible in gtkperf benchmark results. In my case gtkperf run takes 3 times longer (I'll attach detailed results). I have an ATI Technologies Inc RS690M [Radeon X1200 Series]. Version-Release number of selected component (if applicable): kernel-2.6.27.5-120.fc10.x86_64 xorg-x11-server-Xorg-1.5.3-5.fc10.x86_64 xorg-x11-drv-ati-6.9.0-55.fc10.x86_64 How reproducible: 100% Steps to Reproduce: 1. Boot normally (with kernel modesetting by default) 2. run gtkperf 3. Repeat the test with 'nomodeset' boot parameter. Actual results: With 'nomodeset' it is significantly faster. Expected results: Kernel mode setting should not negatively affect performance.
Created attachment 324297 [details] gtkperf results with kms enabled
Created attachment 324298 [details] gtkperf results with nomodeset
Created attachment 324299 [details] Xorg.0.log with kms
Created attachment 324300 [details] Xorg.0.log with nomodeset
Created attachment 324301 [details] dmesg with kms
Created attachment 324302 [details] dmesg with nomodeset
I noticed that both /proc/mtrr contents and contents of /sys/kernel/debug/x86/pat_memtype_list change when 'nomodeset' is added. I suspected it might explain the performance difference, but neither adding 'nopat' nor modifying /proc/mtrr manually had any measurable effect on performance.
Possibly duplicate of bug 467506 ?
really gtkperf hasn't gotten any slower except for the lines test. Nobody draws lines like the lines test in normal desktop use cases. I can probably spend the time making lines go faster but it doesn't provide any gains to the apps we normally use. So unless you have real feedback on what apps are acting slower and how slow they are I'll probably not be able to do anything with this. please reopen if you have a real use case that is noticably slower, or not on bug 467506.