Announcement

Collapse
No announcement yet.

R6xx/R7XX kernel 2.6.33 module performance hacks

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Obscene_CNN
    replied
    Xming and I will probably do that once we get the git server up.

    I have found that on my computer I have maxed out my video performance with Torcs racing sim on the Frosa track. I no longer get better FPS with further improvements. But I get a more responsive game. On tracks with simpler scenery the FPS seems to be still going up.

    I attribute these changes to my GPU being maxed out for processing power on the Frosa track but I have more CPU cycles to do the simulation with. On tracks with simpler scenery I'm not saturating the GPU and with the CPU cycles saved I can feed the GPU more data.

    I'm currently out of ideas on how to improve the KMS code path much more. I have however improved the non KMS code more and there also should be some slight improvements to r100, 300, and 500 chips do to these changes.

    I'll probably have another set of patches by the this weekend.

    Leave a comment:


  • that guy
    replied
    If you set up a git repo, you may want to poke the Zen Kernel guys, as I guess Zen Kernel users are your target demographic which I assume is users looking for that extra 1% or 2% in performance.

    Leave a comment:


  • Obscene_CNN
    replied
    As I understand it xming is working on setting up a public git repo

    Leave a comment:


  • elanthis
    replied
    Is there any sane reason these patches are being distributed as they are instead of as commits in a cloned git repo? github is free, gitorious is free, sourceforge git is free, google code git is free...

    Leave a comment:


  • xming
    replied
    New patches from Obscene_CNN, applies here and working. Not much more performance here YMMV.

    Leave a comment:


  • Obscene_CNN
    replied
    New hack coming, mainly for faster non KMS .

    Leave a comment:


  • xming
    replied
    Just got the patches, tested and working, although I don't see (subjectively) performance difference.

    patch against .33-rc or radeon-testing

    Patch against 20100113 (the previous patches from obscene_cnn).

    Leave a comment:


  • Obscene_CNN
    replied
    Originally posted by darkbasic View Post
    Nexuiz timedemo - demo1
    2560x1600, KMS, medium effects, Radeon HD3870
    without the patch: min avg max 14/28/56
    with the patch: min avg max 14/28/56
    Updated KMS patch is coming.... It has more optimizations done to the KMS code. I have given the patches to xming and he will test them and release them probably tomorrow.

    Leave a comment:


  • darkbasic
    replied
    Nexuiz timedemo - demo1
    2560x1600, KMS, medium effects, Radeon HD3870
    without the patch: min avg max 14/28/56
    with the patch: min avg max 14/28/56

    Leave a comment:


  • xming
    replied
    tested your new set of patches, work great. I had previously performance problem with glxgears, but found the problem, it was caused by kwin desktop effect, apparently that ate so much performance compared to compiz.

    Now w/o your patches glxgears runs at 2000~2100 fps, with it it runs at 2300~2370. Biggest improvement comes from ut2004, unplayable framerate to somehow more playable, I need to play more maps to judge the difference.

    I have produced a bz2 which contains the 3 patchs in one, a bit easier to download and apply (cleanly to 2.6.32-drm-radeon-testing) here. Obscene_CNN, I hope you don't mind.
    Last edited by xming; 15 January 2010, 07:53 AM.

    Leave a comment:

Working...
X