From: Christian Wohlschlager (Christian.Wohlschlager_at_jku.at)
Date: Fri Oct 23 2009 - 00:20:59 CDT

** Reply Requested When Convenient **

Hello John !
 
Thank you very much for the quick Answere ,Yes there is some inconsistency only n004 got driver version 190.xxx all the others got version below ( 175.xxx) , i try to update them also ( Murfy Law No:.1 Never change a running System) :-)
 
mfg
 
Christian
 
 
-- Johannes Kepler Universitaet
-- Christian Wohlschlager
-- Altenbergerstr.69
-- 4040 Linz Austria
-- Tel: +43 732 2468 8595
- Mail:christian.wohlschlager_at_jku.at

>>> John Stone <johns_at_ks.uiuc.edu> 22.10.2009 18:04 >>>
Hi,
  The fact that you're able to see CUDA on n004:0 tells me that there's some
kind of hardware or software inconsistency on the nodes of your machine.
Does each node actually have a GPU, or do they have GPUs clumped together
in a smaller number of nodes? You will need to verify that the NVIDIA
driver running on every single node is 190.x. You can easily check with:
  cat /proc/driver/nvidia/version

If the nodes all have the correct driver version, then the next thing to
check is to make sure the driver is actually loaded and initialized on each
node, I've attached a simple "startcuda" script to this email that you can
use to force the video driver to be loaded and to initial the GPU(s) on
a machine, even if X isn't currently running. You can use this to enable
CUDA on cluster nodes that aren't running X. You may need to tweak the
script if your Linux distro is very different from the usual, but you'll
see what it's doing and it's pretty easy to follow.

Give these things a shot and let me know how it works for you.

I've attached the startup script to this email.

Cheers,
  John Stone
  vmd_at_ks.uiuc.edu

On Thu, Oct 22, 2009 at 08:56:18AM +0100, Christian Wohlschlager wrote:
> Hy
> i have a cluster (sgi virtu) with 8 nodes (n001,n002,n003...n008) a Head
> node named virtu. this cluster got 8 Nvidia
> with each having Nvidia 5600 Quatro installed only one =n004:0 on has no
> graphik window attached so is free for GPU computing ,the graphic driver
> is 190.xxx , if i start the process on virtu i cannot find the GPU ( -->
> No Cuda acceleration device found) .
> How can i specify the Nvidia card it should look for ?
>
> (If i start vmd on n004 it can find a acceleration device and it is
> working)
>
> mfg
>
> Christian
>
> -- Johannes Kepler Universitaet
> -- Christian Wohlschlager
> -- Altenbergerstr.69
> -- 4040 Linz Austria
> -- Tel: +43 732 2468 8595
> - Mail:christian.wohlschlager_at_jku.at

-- 
NIH Resource for Macromolecular Modeling and Bioinformatics
Beckman Institute for Advanced Science and Technology
University of Illinois, 405 N. Mathews Ave, Urbana, IL 61801
Email: johns_at_ks.uiuc.edu                 Phone: 217-244-3349
  WWW: http://www.ks.uiuc.edu/~johns/      Fax: 217-244-6078