[torqueusers] Re: [Mauiusers] GPUs as a calculation elements in cluster.

James A. Peltier jpeltier at cs.sfu.ca
Sun Jul 20 23:41:29 MDT 2008


On Thu, 17 Jul 2008, Egor Tur wrote:

> Hi folk.
>
> What we have - each node in cluster has 2 GPUs and 8 cores (2 quad CPU).
> What we want - create different queues (1) for GPU jobs, (2) for pure CPU jobs and (3) for mixed GPU+CPU jobs.
> Case (1) means that all jobs use GPU on all nodes.
> Case (2) means that all jobs use only CPU on all nodes.
> Case (3) means that all jobs use as GPU as CPU (as threads) on each node.
>
> How we can organise these by torque and maui?
> For CPU jobs we must create queue which have only 6 CPU per node. (other 2 CPU are used by GPU)
> For GPU jobs we must create queue which have 2 CPU per node.
> And we what to have queue for jobs which use one GPU and 4 CPU per node.
>
> If is it possible? If yes, how we can do this? Who have experience with the same problem?
>

I'm interested in this as well, as we will be purchasing a GPU based 
cluster solution in the not so distant future.  My first incline is to 
setup each not with a consumable resource.  That resource could be defined 
then in the job submission as

qsub -l nodes=1:gpus=2

or some such thing.  I too am looking for input though :)

-- 
James A. Peltier
Systems Analyst (FASNet), VIVARIUM Technical Director
Simon Fraser University - Burnaby Campus
Phone   : 778-782-6573
Fax     : 778-782-3045
Mobile  : 778-840-6434
E-Mail  : jpeltier at sfu.ca
Website : http://www.fas.sfu.ca | http://vivarium.cs.sfu.ca
MSN     : subatomic_spam at hotmail.com


More information about the torqueusers mailing list