Am Friday 07 January 2011 schrieb oliverthered:
> I'm getting an issue what appears to be GPU memory issues, specifically the
> memory seems to be getting eaten up quite quickly even though I've got
> loads of system ram and not a huge number of application windows or apps
> running. I belive my graphics card has 512MB or ram or maybe 1GB but it's
> a GeForce 7900 GS.
"nvidia" or "nouveau"?

> I've changed over to none oxygen version of everything (as that was pointed
> to as the culprit for some leakyness), but closing down windows seems to
> free up the ram and I can run things like projectM full screen again and
> with no performance issues.
Can you actually confirm that the RAM hunger is distributed evenly among your 
running applications ("top") and not everything is just sucked by only one 
(X11?) process?

Usually graphical stuff (including and good god esp. opengl) ends up in your 
VRAM (ie. on the GPU), you can query the used X11 resources by "xrestop"

in doubt: ask your distro whether they run anything (everything...) by 
"--graphicssystem raster" and then tell them to stop doing this ;-)
 
> I would have expected that opengl should put stuff into system memory as
> it's ment to do it's own memory mangement, at least for some stff etc...
yes, as last resort - but your GPU can carry some windows before mapping to 
sysram.
please notice that ARGB enabled windows will take 33% more memory than 
"ordinary" ones. konsole and most plasma-desktop windows use ARGB and we 
recently figured that apparently gdk driven GL clients (like at least eg. 
cairo-dock) will make all Qt clients (or rather everything?) use ARGB 
drawables

the nvidia driver has however it's very own pitfall.
call "nvidia-settgins -a PixmapCache=0; nvidia-settgins -a PixmapCache=1" from 
time to time (every 30 minutes and right after your desktop is loaded) to 
improve performance...

sidenote:
consider to split up the mail, strip it a bit ... more... and resend it
=P

Cheers,
Thomas
 
>> Visit http://mail.kde.org/mailman/listinfo/kde-devel#unsub to unsubscribe <<

Reply via email to