First of all: It is Linux, and Suse ... (sorry for using Suse ...)

Today I (somehow accidently) upgraded my machine and since now CUDA is working.

So far fine. The first GPU-WU is already waiting for validation and is really faster than the CPU.

However, the problem ist that my machine (actuall only X) is really, really slow when boinc is running. As I found out with the utility "top", the programm X is using something between 30 and 60% of one of the CPU's. If I stop boinc, X just uses 1-5% of CPU and the machine is really fast.

Now, there is an option (in both the local boinc manager and on the website) to stop using the GPU while a user (or my cat if i had one) is moving the mouse or pressing any key. But this has no effect on my machine, the only way to make it usable again is to stop *all* boinc and all 8 CPU's.

I start boinc with a special user "boinc" from cron with an entry like this:

First of all: It is Linux, and Suse ... (sorry for using Suse ...)

However, the problem ist that my machine (actuall only X) is really, really slow when boinc is running. As I found out with the utility "top", the programm X is using something between 30 and 60% of one of the CPU's. If I stop boinc, X just uses 1-5% of CPU and the machine is really fast.

Now, there is an option (in both the local boinc manager and on the website) to stop using the GPU while a user (or my cat if i had one) is moving the mouse or pressing any key. But this has no effect on my machine, the only way to make it usable again is to stop *all* boinc and all 8 CPU's.

I start boinc with a special user "boinc" from cron with an entry like this:

# crontab -u boinc -l
# DO NOT EDIT THIS FILE - edit the master and reinstall.
# (/tmp/crontab.XXXXn8gDwH installed on Sun Dec 19 17:26:38 2010)
# (Cron version V5.0 -- $Id: crontab.c,v 1.12 2004/01/23 18:56:42 vixie Exp $)
@reboot cd ~/BOINC;nohup ./run_client --daemon

Of course, I do not use the machine with that account (a shell like /bin/false is not really usable *g*). And for h[i|y]st[o|e]rical reasons I do not even start X when the machine boots.

[edit]
What hacks do I need to make the X-Server responding faster? Or what changes to the configuration is needed for boinc to detect mouse-moves/keyboard pressing?
[/edit]

Now some infos ...
My computer:
http://einstein.phys.uwm.edu/show_host_detail.php?hostid=1967526

the nvidia module has version 260.19.44 (from March 7, 2011)

the kernel is 2.6.38-rc7-2-default (causing nvidia 260.1936 from January 21, 2011 not to work ... for whatever reasons).

boinc:
BOINC client version 6.10.17 for x86_64-pc-linux-gnu

First of all: It is Linux, and Suse ... (sorry for using Suse ...)

I shouldn't answer you, cause of that excuse. ;)

You don't need that setting, X should always react very fast(It does here), cause all apps run in the background. So it must be something else that is slowing down your desktop.
You can only run the CUDA apps if the driver is loaded through X(logged in!) or if the CUDA module is loaded at boot time see here.

Install the new Beta driver from here(270.26)
Then you can use the BRP3cuda32NV270 app, that will use no full CPU, but about 33% of one CPU.

the kernel is 2.6.38-rc7-2-default (causing nvidia 260.1936 from January 21, 2011 not to work ... for whatever reasons).

Why do you use a kernel without support from OpenSUSE? Installing the new Version 11.4 will bring the kernel 2.6.37 to you. Is this one too old? With all kernels from OpenSUSE you will get drivers from Nvidia that will work.

You could also try to start boinc with the option --check_all_logins, if you want to use mouse/keyboard detection.

Last not least, your GPU has 512MB RAM available. The CUDA apps usually use 300MB, but as otheres reported they can use up to 400MB. This will leave only a small amount of memory for X. If you start boinc(with X) and change user afterwards, it might be very tight. Better use the advice from the Nvidia forum and create an init-script that starts BOINC with the user boinc.

Check the amount of free GPU memory with the command 'nvidia-smi -q'.

Seems that it is loaded at boot time, since there is a file /etc/modprobe.d/50-nvidia.conf. I rebootet the machine after installing the nvidia driver just to be sure, that everything really works (not because it was needed).

From the nvidia page ...
270.26 February 21, 2011
260.1944 March 7, 2011

260.1944 is the newest (considering at the date)

the kernel is 2.6.38-rc7-2-default (causing nvidia 260.1936 from January 21, 2011 not to work ... for whatever reasons).

Why do you use a kernel without support from OpenSUSE? Installing the new Version 11.4 will bring the kernel 2.6.37 to you. Is this one too old? With all kernels from OpenSUSE you will get drivers from Nvidia that will work.

This is the Suse kernel. For whatever reasons (and yes, this was accidently) I upgraded to 11.5 and not to 11.3 (which was was my plan).

This is a Beta version, that might cause problems.

You could also try to start boinc with the option --check_all_logins, if you want to use mouse/keyboard detection.

# ./boinc --help
...
Usage: ./boinc [options]
...
--check_all_logins for idle detection, check remote logins too

So, this options checks remote logins. Although my computer is placed under my desk, I think this is not what "remote login" means :-)

At least I interpret that line as a mean to additionally check remote logins, and not to check activity at all (including remote logins).

Last not least, your GPU has 512MB RAM available. The CUDA apps usually use 300MB, but as otheres reported they can use up to 400MB. This will leave only a small amount of memory for X. If you start boinc(with X) and change user afterwards, it might be very tight. Better use the advice from the Nvidia forum and create an init-script that starts BOINC with the user boinc.

Check the amount of free GPU memory with the command 'nvidia-smi -q'.

I tried this. Thats what it says ...

nvidia-smi -a

45% of memory seems to be okay.

But my question was: How is keyboard/mouse activity detected? I can live with a system which reacts slow when I start using it and is fast again afterwards. I cannot live with a system which reacts like an old 386 ...

Or, if that does not work: How can I disable CUDA?

...You also could upgrade to boinc 5.10.58. maybe this will solve the problem...

LOL, yes.
____________

Aha! USB might explain it.

6.12 will hopefully have a fix for that.
- client, linux: use X server for idle detection if available.
____________
Jord

-The BOINC FAQ Service
- BOINC 7.0 FAQ

I used to be an adventurer like you. Then I took an arrow in the knee...

Now things are getting strange.

Suddenly I see "Stopped, User is active"¹. The only thing I changed was stopping (or rather suspending) Boinc in its menu. So it seems to work "sometimes".

<Conspiracy theory>EDP really means Experimental Data Processing.</Conspiracy theory>

But for whatever reason I got a lot of memory allocation errors.
http://einstein.phys.uwm.edu/results.php?hostid=1967526&offset=0&show_names=0&state=5
All of them are CUDA WUs, the regular CPU computes fine. And no, there is no screen saver, just blank screen.

Still CUDA-Memory errors? However, this is was nvidia-smi says: