Different performance with different user accounts?

Maciej Cencora m.cencora at gmail.com
Fri Dec 11 16:16:11 PST 2009


Dnia piątek, 11 grudnia 2009 o 23:52:25 Ancoron Luciferis napisał(a):
> Michel Dänzer wrote:
> > On Thu, 2009-12-10 at 00:45 +0100, Ancoron Luciferis wrote:
> >> I have a machine with two user accounts under Kubuntu 9.10 and the
> >> performance difference I am experiencing is measurable. The first user
> >> account (set up during installation time) works good. Glxgears is around
> >> 4000 FPS, video playback is fine, compositing is fine (especially the
> >> desktop cube is fast as hell and really smooth).
> >>
> >> On the other hand there is a second user account (this one is not in the
> >> admin group but basic setup is the same). In that user account I get
> >> only around 2700 FPS, compositing works fine (the cube is still that
> >> fast) but moving windows around is dead slow and video playback is NOT
> >> "tear-free". The rather strange thing is that it doesn't depend on video
> >> source frame size or codec or scaled frame size, the playback is just
> >> dropping frames especially when there is a lot of movement.
> >>
> >> Turning off the compositing results in a fast user experience but since
> >> it is fast with compositing for the first user I expected it to be the
> >> same for the second one.
> >>
> >> CPU usage is not the problem as the quad-core machine is idling around
> >> between 70% and 90% although Xorg is between 30 and 50% and mplayer
> >> between 1% and 26% (peeks up to 47%) when watching a movie.
> >>
> >> System:
> >> - - AMD Athlon II X4 620
> >> - - 4GB RAM
> >> - - Radeon X1950XTX
> >> - - Dell 3007WFP-HC 30" @ 2560x1600
> >>
> >> This doesn't make sense for me at all.
> >>
> >> Can someone shed some light on this?
> >
> > Have you tried comparing e.g. the glxinfo output between the two
> > accounts?
> 
> 'glxinfo -v' yields the exact same results under both user accounts.
> 
> I found out what caused the dramatic performance lag: amor
> 
> The second user is running the small cat 'neko'. When disabled, the user
> experience is the same as it is with the first user.
> 
> glxgears with amor running:
> 13931 frames in 5.0 seconds = 2786.095 FPS
> 13475 frames in 5.0 seconds = 2694.940 FPS
> 13110 frames in 5.0 seconds = 2616.737 FPS
> 13515 frames in 5.0 seconds = 2702.943 FPS
> 16012 frames in 5.0 seconds = 3202.190 FPS
> 14027 frames in 5.0 seconds = 2795.201 FPS
> 15920 frames in 5.0 seconds = 3180.834 FPS
> 13449 frames in 5.0 seconds = 2689.747 FPS
> 
> glxgears without amor:
> 19291 frames in 5.0 seconds = 3858.148 FPS
> 20093 frames in 5.0 seconds = 4018.454 FPS
> 20013 frames in 5.0 seconds = 4002.501 FPS
> 20895 frames in 5.0 seconds = 4178.815 FPS
> 20834 frames in 5.0 seconds = 4166.731 FPS
> 20168 frames in 5.0 seconds = 4033.281 FPS
> 20572 frames in 5.0 seconds = 4114.335 FPS
> 20308 frames in 5.0 seconds = 4061.359 FPS
> 
> Also the video playback is smooth as it should be (even with 1920x1080
> content) and moving windows around is as fast as with the first user
> account.
> 
> But this is still a bit strange to me, as the CPU usage for that amor is
> nearly 0% all the time. So it must somehow interfere the rendering
> pipeline. Could this still be an issue in the driver?
> 
> 
> Thanx,
> 
> Ancoron
> 
> _______________________________________________
> xorg-driver-ati mailing list
> xorg-driver-ati at lists.x.org
> http://lists.x.org/mailman/listinfo/xorg-driver-ati
> 

Make sure you have CPU frequency scaling disabled when running these tests. 

Regards,
Maciej Cencora



More information about the xorg-driver-ati mailing list