[torquedev] GPU Cluster Support

[Craig West]

On 03/08/10 23:22, Chris Reynolds wrote:
> I know PBS Pro has support for dealing with multiple GPUs in one node.
> Has or will this be added to Torque? We're about to construct a
> reasonable sized GPU cluster, so this functionality would be essential.

I'd like to be added to the list of users wanting this support.

Ideally it would be good to be able to restrict access to GPUs in a 
similar way to cpusets. The reason for wanting this is that some 
applications automatically detect the GPUs and use all they can find, if 
the user only requested one (and the system has more than one), then 
when another job is launched (also requesting a GPU) it will die (or run 
slowly) as the GPU it was to be allocated is already in use.

Note: I believe this is possible with cgroups.

See the archive for a previous discussion on this topic.

http://www.clusterresources.com/pipermail/torquedev/2009-June/001592.html

Cheers,
Craig.

-- 
Craig West                   Systems Manager
Victorian Partnership for Advanced Computing
110 Victoria Street, Carlton South  VIC 3053
P: +61 3 9925 4751         E: cwest at vpac.org
                          http://www.vpac.org