NVIDIA 177.67 Display Driver

[kriko]

Indeed, by following posted guide performance in KDE4 does improve. Try it.

[deanjo]

I've got a GF8600GT, and 2D has so far been OK, but not great. I'm using standard Gnome/Ubuntu Hardy.

By following this guide:

http://www.nvnews.net/vbulletin/showthread.php?t=118088

and installing the latest nvidia BETA driver, I'm seeing some amazing 2D performance improvements and haven't noticed any problems yet. Cool !

Ya, the feedback has been looking very good so far.

http://www.nvnews.net/vbulletin/showthread.php?t=118073

[apaige]

Excellent performance with both GTK (redrawing windows is finally instantaneous) and Firefox 3 (no more slowdowns with images getting displayed or resized, digg.com bug gone, scrolling and animated scripts are smooth). That's with my 8600GT. Looks like they did their job well.

The only remaining issue I'm witnessing concerns urxvt with a background image. It still takes a couple seconds to appear or to be resized.

[Fuchs]

Excellent performance with both GTK (redrawing windows is finally instantaneous) and Firefox 3 (no more slowdowns with images getting displayed or resized, digg.com bug gone, scrolling and animated scripts are smooth). That's with my 8600GT. Looks like they did their job well.

The only remaining issue I'm witnessing concerns urxvt with a background image. It still takes a couple seconds to appear or to be resized.

Please check whether this only happens with IPP=2. If so: submit a bug report at the nvnews forums. nVidia devs asked for as much feedback as possible, so they can fix bugs in IPP=2 so they can make this the new default.

You might be interested in using real transparency in urxvt as well, which performs good here. An example configuration would be

urxvt --geometry 165x55 -ut -sr -ls +st -fg white -tint darkblue -sl 1000000 --scrollColor darkblue --troughColor darkblue --depth 32 -bg rgba:0000/0000/3900/cccc

which starts an urxvt with a blue tinted transparent background, so you get your desktop image as background.

hth

Fuchs

[apaige]

which starts an urxvt with a blue tinted transparent background, so you get your desktop image as background.

That's what I'm doing already, actually. And that's not "real" transparency, since it doesn't show windows behind it, which is exactly the effect I want. Real transparency with xcompmgr and transset works very well, but then the focused terminal becomes unreadable whenever there a second terminal behind it (compositing also poses the issue of vertical synching when playing videos with XVideo).

The problem occurs with InitialPixmapPlacement=2 only indeed.

[Fuchs]

That's what I'm doing already, actually. And that's not "real" transparency, since it doesn't show windows behind it, which is exactly the effect I want.

In fact it is, as soon as you add --depth 32 -bg rgba:rrrr/gggg/bbbb/aaaa

however, this requires, afair, a composited environment (as you get with xcompmgr or by using compiz.) Transset should not be needed.

Please look at

http://www.fuchsnet.ch/stack/urxvt.png

or

http://www.fuchsnet.ch/stack/uxvt.png

which clearly demonstrates it.
I did not use transset or any other tricks.

The problem occurs with InitialPixmapPlacement=2 only indeed.

please make sure to inform nvidia about that by sending a nvidia-bug-report.log (created with nvidia-bug-report.sh) and a how to reproduce description to either the [email protected] address or post it on nvnews.net.

So nvidia engineers can try to fix it.

Fuchs

[apaige]

In fact it is, as soon as you add --depth 32 -bg rgba:rrrr/gggg/bbbb/aaaa

however, this requires, afair, a composited environment (as you get with xcompmgr or by using compiz.) Transset should not be needed.

Ah, indeed. Thanks a lot for the tip, it's much more convenient than using transset.

please make sure to inform nvidia about that

Will do.

[Dragoran]

177.68 (beta) is out:

http://www.nvnews.net/vbulletin/showthread.php?t=118244

[apaige]

Update on my rxvt-unicode with fake transparency performance issue: I recompiled it with only the features I wanted, which does NOT include Xft/freetype. Now it's as fast with InitialPixmapPlacement=2 as with InitialPixmapPlacement=1. Not sure what caused the issue.