nV News Forums

 
 

nV News Forums (http://www.nvnews.net/vbulletin/index.php)
-   NVIDIA Linux (http://www.nvnews.net/vbulletin/forumdisplay.php?f=14)
-   -   nvidia 256.52 xorg-server 1.9.0 performance regression (http://www.nvnews.net/vbulletin/showthread.php?t=154563)

ioni 08-29-10 07:07 AM

nvidia 256.52 xorg-server 1.9.0 performance regression
 
1 Attachment(s)
Hi,
i was using nvidia 256.44 which doesn't have xorg 1.9 support with IgnoreABI true and i've noticed performance regressions when scaling windows, when text is scrolling in gnome-terminal.

I was thinking that's because 256.44 didn't had proper support and now that 256.52 is out and has xorg-server 1.9 support, it has the same issues(ignoreabi was removed).

i'm attaching the current xorg.0.log.

using gnome with compositing on in metacity

gradinaruvasile 08-29-10 07:16 AM

Re: nvidia 256.52 xorg-server 1.9.0 performance regression
 
Metacity uses software compositing that is very slow (compiz uses opengl that is fast).

ioni 08-29-10 07:22 AM

Re: nvidia 256.52 xorg-server 1.9.0 performance regression
 
Quote:

Originally Posted by gradinaruvasile (Post 2309240)
Metacity uses software compositing that is very slow (compiz uses opengl that is fast).

not very helpful.
my thread should be consider as a bug report.

the point is that 256.44 with xorg-server 1.8.1.902 works with my current setup and 256.52 with xorg-server 1.9.0 has huge performance regressions even when i disable compositing

gradinaruvasile 08-29-10 07:27 AM

Re: nvidia 256.52 xorg-server 1.9.0 performance regression
 
Have you tried disabling the compositing extension from xorg.conf? Does that make a difference?

Just add

Section "Extensions"
Option "Composite" "Disable"
EndSection

to the end of xorg.conf then restart x.

noclaf 08-29-10 09:02 AM

Re: nvidia 256.52 xorg-server 1.9.0 performance regression
 
1 Attachment(s)
+1 for this. Most applications are okay, but ncurses apps (which I use heavily) are very slow to redraw. There's a noticeable top to bottom screen wipe. Plain terminal sessions (I use urxvt) appear to be unaffected. glxgears shows no choppiness and gtk applications such as Chromium seem to be functioning normally as well.

I use xcompmgr for transparency effects only. Disabling it and/or the compositing extensions has no noticeable effect.

I've attached a full bug report log.

gqman69 08-29-10 11:03 AM

Re: nvidia 256.52 xorg-server 1.9.0 performance regression
 
+1

Same.

adamk75 08-29-10 11:18 AM

Re: nvidia 256.52 xorg-server 1.9.0 performance regression
 
Quote:

Originally Posted by gradinaruvasile (Post 2309240)
Metacity uses software compositing that is very slow (compiz uses opengl that is fast).

That is not true... Metacity uses RENDER compositing, which can certainly be hardware accelerated and quite fast depending on the driver. .

Adam

gradinaruvasile 08-29-10 01:41 PM

Re: nvidia 256.52 xorg-server 1.9.0 performance regression
 
Quote:

Originally Posted by adamk75 (Post 2309317)
That is not true... Metacity uses RENDER compositing, which can certainly be hardware accelerated and quite fast depending on the driver. .

Adam

Which driver? I use only nvidia cards and whenevr i tried the metacity compositing, moving around a single little window spiked the CPU to 50-60%...

ShiningArcanine 08-29-10 04:14 PM

Re: nvidia 256.52 xorg-server 1.9.0 performance regression
 
I have KDE 4.4.5 and CPU usage spikes to about 20% from 2% when I move a window around.

This is with xorg-server 1.8 and nvidia-drivers 256.52.

Licaon 08-29-10 05:19 PM

Re: nvidia 256.52 xorg-server 1.9.0 performance regression
 
Quote:

Originally Posted by ShiningArcanine (Post 2309433)
I have KDE 4.4.5 and CPU usage spikes to about 20% from 2% when I move a window around.

This is with xorg-server 1.8 and nvidia-drivers 256.52.

with composition enabled in Kwin a window move yields 10% CPU usage on my sistem: 5% kwin and 5% Xorg :(
it's not the end of the world but eh...

zacbarton 08-30-10 03:29 AM

Re: nvidia 256.52 xorg-server 1.9.0 performance regression
 
im seeing the exactly same thing

using GtkPerf the 2 text tests go so slow i can see the text being drawn

GtkEntry - time: 0.05
GtkComboBox - time: 0.86
GtkComboBoxEntry - time: 0.68
GtkSpinButton - time: 0.14
GtkProgressBar - time: 0.25
GtkToggleButton - time: 0.29
GtkCheckButton - time: 0.13
GtkRadioButton - time: 0.33
GtkTextView - Add text - time: 3.36
GtkTextView - Scroll - time: 2.86
GtkDrawingArea - Lines - time: 0.50
GtkDrawingArea - Circles - time: 0.58
GtkDrawingArea - Text - time: 11.59
GtkDrawingArea - Pixbufs - time: 0.61

dae 08-30-10 12:04 PM

Re: nvidia 256.52 xorg-server 1.9.0 performance regression
 
I'm also seeing a noticable slowdown in text rendering with xorg-server 1.9.0 (using nvidia 256.52). Gtkperf output:

GtkEntry - time: 0.24
GtkComboBox - time: 1.06
GtkComboBoxEntry - time: 1.11
GtkSpinButton - time: 0.56
GtkProgressBar - time: 0.74
GtkToggleButton - time: 0.58
GtkCheckButton - time: 0.33
GtkRadioButton - time: 0.43
GtkTextView - Add text - time: 3.16
GtkTextView - Scroll - time: 1.57
GtkDrawingArea - Lines - time: 0.35
GtkDrawingArea - Circles - time: 0.52
GtkDrawingArea - Text - time: 12.89
GtkDrawingArea - Pixbufs - time: 0.58

It appears to be related to anti-aliasing; if I turn off anti-aliasing the performance problem is gone:

GtkEntry - time: 0.22
GtkComboBox - time: 0.87
GtkComboBoxEntry - time: 1.16
GtkSpinButton - time: 0.64
GtkProgressBar - time: 0.37
GtkToggleButton - time: 0.29
GtkCheckButton - time: 0.16
GtkRadioButton - time: 0.15
GtkTextView - Add text - time: 0.46
GtkTextView - Scroll - time: 0.58
GtkDrawingArea - Lines - time: 0.36
GtkDrawingArea - Circles - time: 0.52
GtkDrawingArea - Text - time: 0.18
GtkDrawingArea - Pixbufs - time: 0.16

This is quite interresting since the reason I turned on anti-aliasing a couple of QT and nvidia releases back, was due to non-anti-aliased text being slow...


All times are GMT -5. The time now is 08:53 AM.

Powered by vBulletin® Version 3.7.1
Copyright ©2000 - 2014, Jelsoft Enterprises Ltd.
Copyright 1998 - 2014, nV News.