8800 GTX slow with KDE

[d13f00l]

Originally Posted by kriko

This is nvidia problem. In latest drivers they fixed this issue for 6xxx series.

Are you sure it's not an optimization problem? Maybe one widget stores images in a buffer and reuses if possible, while another redraws each character every time there is an update.

It'll probably get fixed by nvidia, but blaming nvidia 100% doesn't explain why one widget would be better faster than another when both are just drawing text.

[kriko]

Can you provide links to problematic widgets? I'll try it.

[d13f00l]

Originally Posted by kriko

Can you provide links to problematic widgets? I'll try it.

This was ages ago. There was a difference in speed between KWrite and KEdit, just scrolling up and down a lot of subpixel hinted text resulted in one of the programs slowing down drastically and using 100% CP, and the other being totally fine w/ the Nvidia drivers. Internally, they used different QT widgets. I dug into the code, but this was well over a year ago. I don't have any KDE machines anymore to try this on or to look up class names on, unfortunately.

[bugger]

update:
compiled the new 2.6.23-rc2 today using the new CFS. Strange thing: the Compiz benchmark shows now 85 fps.

[kriko]

@bugger: new cfs is greatly improved, though I preffer SD scheduler (-ck patchset). You might give it a try, I find it best for 3d gaming.

[Cadynum]

Have you guys at nvidia being able to replicate it yet?

[Cadynum]

I mixed around a bit with
"nvidia-settings -a InitialPixmapPlacement=x"

And x=2 makes my firefox, xfce terminal and possibly more programs fast.
But it also slows down some programs _very_ much (e.g. gimp).

x=1 and x=0 makes gimp fast, but the programs mentioned above suffers from slowdowns then.

I know i have tired x=2 before (it's mentioned in the beginning of the thread), but it didn't work then.
Now i run ck+cfs+some other patches, gcc 4.2 and loads of news stuff.
I don't know why it works now, but I'm happy it does!

[leeman2000]

nvidia-settings -a InitialPixmapPlacement=2 gives me a lot of garbage at the window decorations (metacity) can anybody confirm ? By the way doesn't seem to get things working much faster, sorry

[Cadynum]

Originally Posted by leeman2000

nvidia-settings -a InitialPixmapPlacement=2 gives me a lot of garbage at the window decorations (metacity) can anybody confirm ? By the way doesn't seem to get things working much faster, sorry

Sorry, I'm using e17 so i can't confirm.
The pixmap solution seems quite bad anyway. Some things get faster, and some things gets slower. Swifting between them when alternating programs is what I'm doing now.

[Nephyrin]

I'm still having wierd 8800 problems.

I'm currently using 2.6.22 CK patchset and the .11 drivers

On the newest compiz-fusion SVN (though the same problems have existed since older beryl builds) dragging windows around is perfectly smooth, but doing things like minimizing, creating windows, etc causes stuttering. For instance, when opening firefox there is a noticable ~500ms hang before the window appears. Resizing windows also causes this, i can bring my FPS to 0 easily by rapidly resizing a large window.

Pixmap placement 2 seems to alleviate this somewhat, but causes display corruptions and hangs. I have tried AIGLX false in the xorg file, different driver versions, upgraded to xorg 1.3, etc. The newest drivers and newest xorg both seemed to *slightly* relieve the problem, but it seems like there's a different underlying issue that is unsolved. Attached is my xorg.conf as well as kernel config (both of which I've messed with extensively)

Additionally, everything on my system is compiled with -O2 (Optimization level 2, pretty standard) and -march=nocona (Optimize for nocona (Core 2 base), makes everything run more efficiently and enables SSE3 instructions in compiling). I'm going to try recompiling compiz/xorg and some other things now with these CFLAGS removed and see if that helps -- I would upgrade GCC but 4.2 has some optimization errors that KO'd my laptop, so I'm waiting to see what people have to say about 4.2.1/2

[kriko]

@Nephyrin: this is a known issue with compiz (everyone with any card has this problem). Have you tried beryl?

[Nephyrin]

This problem existed on beryl too. I recently upgraded to compiz-fusion (the successor to beryl - they have re-merged with compiz and are now a plugin package) and the same problem exists, no better or worse.

I also just recompiled xorg-server and all of compiz with -march and -O2 removed from CFLAGS (nvidia kernel module already ignores custom CFLAGS), but it produced no difference