[gentoo-user] Abysmally slow 2D performance using proprietary Nvidia driver on dual Xeon system (EMT64)

[gentoo-user] Abysmally slow 2D performance using proprietary Nvidia driver on dual Xeon system (EMT64)

[Greg Bur]

[-- Attachment #1: Type: text/plain, Size: 2937 bytes --]

Greetings,

I have been trying to fix a somewhat puzzling and annoying problem over the
last couple of months that has me absolutely stumped. I have been unable to
isolate the cause of the problem and it appears to be intermittant at best.
The issue rears its ugly head as follows:

This is on a dual boot workstation with a Supermicro X6DAL-XTG motherboard,
dual 3.0GHz Xeon processors and a GeForce 6600GT video card. The system has
worked splendidly since I purchased it back in June with one exception:
occasionally when rebooting the system to return to Linux from Windows the
2D performance in Gnome is terrible. The problem does not appear immediately
nor does it happen every time. Initially gdmgreeter appears as it should and
I enter my login credentials. The Gnome desktop loads quickly and as I start
to work applications load normally. However, after about 10 minutes the
system load on one processor sharply increases to 100% when performing a
simple task such as clicking on a button in Firefox, launching a new
gnome-terminal window or clicking on the "Applications" menu at the top of
the screen. The response time between clicking and the menu appearing for
instance is anywhere from 2-5 seconds. There does not seem to be any quick
fix for this which is what has me puzzled. If I close all of my
applications, log out and log back in the problem "resets." Rebooting the
system proves ineffective as well. I have tried various driver revisions,
currently using 1.0.6629-r4, and several different kernels to no avail. I
have tried both enabling and disabling the RenderAccel option in
/etc/X11/xorg.conf and that does not appear to make any difference.
Adjusting the NvAGP option does nothing nor does enabling the Composite
option however the system seems more responsive when the Composite option is
enabled. Switching to the "nv" driver resolves the problem and the system
remains snappy with no immediate problems. Occasionally I will begin to get
artifacts on the display with the nv driver and then the screen goes
entirely white but switching to the console using ctrl + alt + F1 usually
resolves this problem. I would appreciate any insight anyone could provide
as to what I am overlooking. I have walked through the Gentoo Nvidia
documentation numerous times and I have consulted the README document
included with the nvidia driver. I am completely stumped.

Currently I am using 2.6.9-gentoo-r9 for a kernel with the
1.0.6629-r4version of the nvidia driver. As I said before I have tried
several kernel
and driver revisions to no avail. As a sidebar, 3D performance seems to be
as it should be which puzzles me further. Again, I appreciate any advice
anyone may offer.

Please see the following links for the last xorg.conf I tried as well as my
kernel config and make.conf

http://pizon.org/xorg.conf
http://pizon.org/kernel-config
http://pizon.org/make.conf

[-- Attachment #2: Type: text/html, Size: 3294 bytes --]

Re: [gentoo-user] Abysmally slow 2D performance using proprietary Nvidia driver on dual Xeon system (EMT64)

[brullo nulla]

> However, after about 10
> minutes the system load on one processor sharply increases to 100% when
> performing a simple task such as clicking on a button in Firefox, launching
> a new gnome-terminal window or clicking on the "Applications" menu at the
> top of the screen.

I have not understood if the system load increases to 100% during the
action of clicking and only during the action or if it goes up and
remains stable.

In the second case you should launch "top" from a shell and see what
process chews up your processor.

m.

-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] Abysmally slow 2D performance using proprietary Nvidia driver on dual Xeon system (EMT64)

[Greg Bur]

[-- Attachment #1: Type: text/plain, Size: 1413 bytes --]

On 11/7/05, brullo nulla <brullonulla@gmail.com> wrote:
>
> > However, after about 10
> > minutes the system load on one processor sharply increases to 100% when
> > performing a simple task such as clicking on a button in Firefox,
> launching
> > a new gnome-terminal window or clicking on the "Applications" menu at
> the
> > top of the screen.
>
> I have not understood if the system load increases to 100% during the
> action of clicking and only during the action or if it goes up and
> remains stable.


My apologies for not making this more clear. The system load spike begins
with the click and it hovers at or near 100% until the new task (opening a
program, displaying a menu, etc) has completed. Even moving a window causes
this to happen and the load only jumps on one processor, the other is idle
or nearly so. When the nvidia driver is working correctly (assuming the
driver is to blame) the load seems to be balanced evenly across both
processors.

In the second case you should launch "top" from a shell and see what
> process chews up your processor.
>
> m.


top is telling me that X is the guilty party. I can renice X to a lower
priority and get some responsiveness back but again, there are times when
everything performs as expected. The problem is intermittant although it
happens more often than not.



--
> gentoo-user@gentoo.org mailing list
>
>

[-- Attachment #2: Type: text/html, Size: 2138 bytes --]

Re: [gentoo-user] Abysmally slow 2D performance using proprietary Nvidia driver on dual Xeon system (EMT64)

[A. Khattri]

On Mon, 7 Nov 2005, Greg Bur wrote:

> My apologies for not making this more clear. The system load spike begins
> with the click and it hovers at or near 100% until the new task (opening a
> program, displaying a menu, etc) has completed. Even moving a window causes
> this to happen and the load only jumps on one processor, the other is idle
> or nearly so. When the nvidia driver is working correctly (assuming the
> driver is to blame) the load seems to be balanced evenly across both
> processors.

What are the specs of the machine? How much RAM when running X? ("free -t
-o -m" in an xterm will tell you). How much free disk space? Dual
processor machine? If so, SMP is enabled in your kernel? Is X using
software rendering?

Many variables here.

-- 

-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] Abysmally slow 2D performance using proprietary Nvidia driver on dual Xeon system (EMT64)

[Greg Bur]

[-- Attachment #1: Type: text/plain, Size: 1474 bytes --]

On 11/7/05, A. Khattri <ajai@bway.net> wrote:
>
> On Mon, 7 Nov 2005, Greg Bur wrote:
>
> > My apologies for not making this more clear. The system load spike
> begins
> > with the click and it hovers at or near 100% until the new task (opening
> a
> > program, displaying a menu, etc) has completed. Even moving a window
> causes
> > this to happen and the load only jumps on one processor, the other is
> idle
> > or nearly so. When the nvidia driver is working correctly (assuming the
> > driver is to blame) the load seems to be balanced evenly across both
> > processors.
>
> What are the specs of the machine? How much RAM when running X? ("free -t
> -o -m" in an xterm will tell you). How much free disk space? Dual
> processor machine? If so, SMP is enabled in your kernel? Is X using
> software rendering?


Dual 3.0Ghz Xeon
2GB RAM
128MB GeForce 6600GT
Audigy 2 soundcard

free -t -o -m output:

gbur@minuteman ~ $ free -t -o -m
total used free shared buffers cached
Mem: 2009 1505 503 0 440 584
Swap: 1953 2 1950
Total: 3962 1508 2454

SMP is enabled in the kernel as well as hyperthreading in the BIOS. As for X
using software rendering, to be completely honest I'm not sure and my guts
are telling me that is what is happening here. I believe I have enabled all
of the appropriate options in the kernel as well as /etc/X11/xorg.conf


Many variables here.
>
> --
>
> --
> gentoo-user@gentoo.org mailing list
>
>

[-- Attachment #2: Type: text/html, Size: 2936 bytes --]

Re: [gentoo-user] Abysmally slow 2D performance using proprietary Nvidia driver on dual Xeon system (EMT64)

[A. Khattri]

On Mon, 7 Nov 2005, Greg Bur wrote:

> Dual 3.0Ghz Xeon
> 2GB RAM
> 128MB GeForce 6600GT
> Audigy 2 soundcard
>
> free -t -o -m output:
>
> gbur@minuteman ~ $ free -t -o -m
> total used free shared buffers cached
> Mem: 2009 1505 503 0 440 584
> Swap: 1953 2 1950
> Total: 3962 1508 2454

According to the output of free it looks like despite having 2Gb of RAM,
your machine is swapping to disk. That will slow down your machine too.


-- 

-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] Abysmally slow 2D performance using proprietary Nvidia driver on dual Xeon system (EMT64)

[Greg Bur]

[-- Attachment #1: Type: text/plain, Size: 1145 bytes --]

On 11/8/05, A. Khattri <ajai@bway.net> wrote:
>
> On Mon, 7 Nov 2005, Greg Bur wrote:
>
> > Dual 3.0Ghz Xeon
> > 2GB RAM
> > 128MB GeForce 6600GT
> > Audigy 2 soundcard
> >
> > free -t -o -m output:
> >
> > gbur@minuteman ~ $ free -t -o -m
> > total used free shared buffers cached
> > Mem: 2009 1505 503 0 440 584
> > Swap: 1953 2 1950
> > Total: 3962 1508 2454
>
> According to the output of free it looks like despite having 2Gb of RAM,
> your machine is swapping to disk. That will slow down your machine too.


It definitely is not an issue with swapping to disk. For some reason X is
causing a high CPU load and it acts like its only using one processor. If I
switch back to the open source driver the load seems to be balanced across
all "4" processors. To make things even more interesting if I open glxgears
while using the Nvidia driver the problem can be temporarily alleviated. The
CPU load drops back to normal and the system is much more responsive. The
open source driver is working fine for now but I sure would like to know
what it is that I'm doing to cause the Nvidia driver to perform so poorly.

[-- Attachment #2: Type: text/html, Size: 1508 bytes --]

Re: [gentoo-user] Abysmally slow 2D performance using proprietary Nvidia driver on dual Xeon system (EMT64)

[A. Khattri]

On Thu, 10 Nov 2005, Greg Bur wrote:

> It definitely is not an issue with swapping to disk. For some reason X is
> causing a high CPU load and it acts like its only using one processor. If I
> switch back to the open source driver the load seems to be balanced across
> all "4" processors. To make things even more interesting if I open glxgears
> while using the Nvidia driver the problem can be temporarily alleviated. The
> CPU load drops back to normal and the system is much more responsive. The
> open source driver is working fine for now but I sure would like to know
> what it is that I'm doing to cause the Nvidia driver to perform so poorly.

Maybe the Nvidia drivers dont play well with threading and/or SMP?


-- 

-- 
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] Abysmally slow 2D performance using proprietary Nvidia driver on dual Xeon system (EMT64)

[Greg Bur]

[-- Attachment #1: Type: text/plain, Size: 1380 bytes --]

On 11/10/05, A. Khattri <ajai@bway.net> wrote:
>
> On Thu, 10 Nov 2005, Greg Bur wrote:
>
> > It definitely is not an issue with swapping to disk. For some reason X
> is
> > causing a high CPU load and it acts like its only using one processor.
> If I
> > switch back to the open source driver the load seems to be balanced
> across
> > all "4" processors. To make things even more interesting if I open
> glxgears
> > while using the Nvidia driver the problem can be temporarily alleviated.
> The
> > CPU load drops back to normal and the system is much more responsive.
> The
> > open source driver is working fine for now but I sure would like to know
> > what it is that I'm doing to cause the Nvidia driver to perform so
> poorly.
>
> Maybe the Nvidia drivers dont play well with threading and/or SMP?


That's my guess that this point. If I get the inspiration tonight I'll try
disabling hyperthreading in the BIOS and also recompiling the kernel without
SMP support. What has me beating my head against the wall so much is not the
poor performance but the fact that one out of every 10 or 15 attempts the
driver works as expected. I suppose if something positive is to be made of
this it is that I will be more reluctant to play any games because I will
have to reboot into Windows.

--
>
> --
> gentoo-user@gentoo.org mailing list
>
>

[-- Attachment #2: Type: text/html, Size: 1899 bytes --]

Re: [gentoo-user] Abysmally slow 2D performance using proprietary Nvidia driver on dual Xeon system (EMT64)

[Greg Bur]

[-- Attachment #1: Type: text/plain, Size: 1092 bytes --]

On 11/10/05, Greg Bur <greg.bur@gmail.com> wrote:
>
>
> >
> > Maybe the Nvidia drivers dont play well with threading and/or SMP?
>
>
> That's my guess that this point. If I get the inspiration tonight I'll try
> disabling hyperthreading in the BIOS and also recompiling the kernel without
> SMP support. What has me beating my head against the wall so much is not the
> poor performance but the fact that one out of every 10 or 15 attempts the
> driver works as expected. I suppose if something positive is to be made of
> this it is that I will be more reluctant to play any games because I will
> have to reboot into Windows.
>
>
Well I tried without hyperthreading and without SMP and the results only got
worse. Now I am beginning to look at how I compiled things such as glibc and
xorg-x11. I already ran the undo process for prelink and now I think I shall
recompile glibc without nptl support to see if that makes any difference. I
am really treading water here but I have been forced to read up on my
hardware a lot more. I really wish I had taken better notes.

[-- Attachment #2: Type: text/html, Size: 1636 bytes --]

Re: [gentoo-user] Abysmally slow 2D performance using proprietary Nvidia driver on dual Xeon system (EMT64)

[Greg Bur]

On 11/11/05, Greg Bur <greg.bur@gmail.com> wrote:
>  Well I tried without hyperthreading and without SMP and the results only
> got worse.  Now I am beginning to look at how I compiled things such as
> glibc and xorg-x11.  I already ran the undo process for prelink and now I
> think I shall recompile glibc without nptl support to see if that makes any
> difference.  I am really treading water here but I have been forced to read
> up on my hardware a lot more.  I really wish I had taken better notes.

Apparently I have fixed the problem now.  I ended up updating to a
newer kernel and used the ~amd64 nvidia drivers and everything is
working as it should.  I am still unsure what the real problem was
assuming there even was a problem.  At any rate, I appreciate the
advice everyone offered.

-- 
gentoo-user@gentoo.org mailing list