Created attachment 331991 [details] Radeon driver in Fedora 9 vs. Fedora 10 (positive is good) Description of problem: Since upgrading my laptop from F9 to F10 the graphic performance has seemed to have gone down, but I had no easy way of verifying this. I thought that surely there must be a site where I can see performance test numbers for each X and driver release but failed to find one. There must be something like this somewhere, right? Recently I acquired a second laptop of the same model and was able to compare the performance of F9 against F10 and confirm my suspicions. I'm attaching x11perf test results that show that Radeon on F10 is slower than before and VESA is faster than Radeon. Version-Release number of selected component (if applicable): Dell Latitude D600 01:00.0 VGA compatible controller: ATI Technologies Inc Radeon RV250 [Mobility FireGL 9000] (rev 02) xorg-x11-drv-ati-6.10.0-2.fc10.i386 xorg-x11-server-Xorg-1.5.3-6.fc10.i386 kernel-2.6.27.12-170.2.5.fc10.i686 How reproducible: Always Steps to Reproduce: 1. Enter runlevel 2 2. Start /usr/bin/X without /etc/X11/xorg.conf 3. Run x11perf on F10 with Radeon driver 4. Run x11perf on F10 with VESA driver 5. Run x11perf on F9 with Radeon driver 6. Compare results Actual results: In general VESA driver is fastest, F9 Radeon next fastest and F10 Radeon slowest. Expected results: The opposite. Additional info: I this is Xorg.0.log on F10 with Radeon, but can't say if it affects 2D performance: (EE) RADEON(0): Static buffer allocation failed. Disabling DRI. (EE) RADEON(0): At least 33600 kB of video memory needed at this resolution and depth. x11perf tests were run with 1s period and with 1 repeat instead of the standard 5s and 5 repeats to keep the testing time manageable (~20 mins vs. 9 hours). Percentages in the graphs tell how many percent the other driver is faster than the other (-100% = speed of A is 2x of speed B, 0% = speed of A is same as B's, 100% = speed of B is 2x of speed A).
Created attachment 331992 [details] VESA vs. Radeon driver on Fedora 10 (positive = Radeon better)
Created attachment 331993 [details] x11perf raw results for Radeon on F10
Created attachment 331994 [details] x11perf raw results for Radeon on F9
Created attachment 331995 [details] x11perf raw results for VESA on F10
Risto, Could you add full dmesg and /var/log/Xorg.0.log as uncompressed text/plain attachments to this bug ? If possible for both F9 and F10. Thanks --- Fedora Bugzappers volunteer triage team https://fedoraproject.org/wiki/BugZappers
Could you also have a look at the fixes mentioned in bug 473115 and report if they work for you ?
Created attachment 332152 [details] Xorg.0.log for Radeon on F9
Created attachment 332153 [details] Xorg.0.log for Radeon on F10
Created attachment 332155 [details] dmesg for Dell Latitude D600 on F10
(In reply to comment #6) > Could you also have a look at the fixes mentioned in bug 473115 and report if > they work for you ? Thanks for the pointer. I wonder how I missed that bug (maybe because it talks about Compiz and I haven't even dared to try that)... I'll try these settings and run the tests again (hopefully soon): Option "AccelMethod" "EXA" Option "AccelDFS" "True" Option "MigrationHeuristic" "greedy"
Created attachment 332874 [details] Radeon driver vs. Radeon driver with EXA acceleration on F10 The performance is ever worse with EXA enabled.
Created attachment 332875 [details] x11perf raw results for Radeon with EXA on F10
The funny thing is that with EXA the system feels much more responsive and usable. E.g. keeping Alt-Tab pressed refreshed the rectangle of the focused windows only about once per second. With EXA the rectangle rapidly goes from one icon to the other and even the window rectangles get draw. Without EXA pressing RETURN in less(1) affected the screen only after a pause, with EXA screen scrolls immediately. Without EXA horizontally scrolling text in the Update System dialog duplicated the rightmost pixels until the scrolling stopped and the text appeared properly. With EXA scrolling happens correctly. I'm wondering if x11perf actually tests throughput and the problem is more of latency... Is there some other test tools that I could try? Why doesn't X server automatically enable EXA? What is the significance of this error: (EE) RADEON(0): Static buffer allocation failed. Disabling DRI. (EE) RADEON(0): At least 33600 kB of video memory needed at this resolution and depth. Would DRI help in 2D performance? Why does DRI need >20 bytes per pixel?
I added some glxgears performance measurements in bug 473440 which show that performance without DRI is abysmal, but DRI gets enabled only in 16bpp mode.
*** This bug has been marked as a duplicate of bug 473115 ***