You are correct that it doesn't have a direct bearing on PBS. Ross has a
few scripts here:
http://ambermd.org/gpus/#Running
that one could dove-tail together in conjunction with a PBS shell script to
properly stagger or stage jobs.
The other-way is to figure out how to get PBS to automatically assign jobs
to a gpu and thereby que jobs when a gpu (via gpu-id is in use).
Otherwise the other-way to slice it is to use the qsub -W flag to manually
stagger the jobs, which IMHO is sub optimal.
On Wed, Sep 19, 2012 at 3:20 PM, Robert Crovella <RCrovella.nvidia.com>wrote:
> I don't think there's anything wrong with your setup at least as far as
> nvidia-smi goes.
>
> I have seen in other scenarios with GTX 680 and 304.43 that nvidia-smi
> reports N/A for utilization.
>
> I don't think that has any direct bearing on getting torque-PBS to treat
> the GPUs separately.
>
> -----Original Message-----
> From: Jonathan Gough [mailto:jonathan.d.gough.gmail.com]
> Sent: Tuesday, September 18, 2012 4:45 PM
> To: AMBER Mailing List
> Subject: [AMBER] nvidia settings cuda jobs with pbs?
>
> I'm running ubuntu 12.04, with a GTX 690 (2 cores)
>
> and I have been trying to set things up so I can use pbs-torque to submit
> jobs to each gpu individually.
>
> I updated torque-PBS to the latest version 2.5.12 (gpu support)
>
> I managed to update the nvidia drivers to v4.304.43
>
> and I have been following instructions here -
> http://ambermd.org/gpus/#Running
>
> That being said when running gpu-info, I can't get a value for Utilization
>
> ====================================================
> Device Model Temperature Utilization
> ====================================================
> 0 GeForce GTX 32 C N/A
> 1 GeForce GTX 33 C N/A
> ====================================================
>
>
> Does anyone have any insight on:
> a. getting the Utilization to work through nvidia-smi? (therefore
> following instructions by ross linked above) b. getting pbs-torque to
> properly queue jobs taking into account the # of gpu's I have?
> c. any other possible solutions to adequately setting up a machine to do
> what I need to do?
>
> Thanks,
> Jonathan
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
> -----------------------------------------------------------------------------------
> This email message is for the sole use of the intended recipient(s) and
> may contain
> confidential information. Any unauthorized review, use, disclosure or
> distribution
> is prohibited. If you are not the intended recipient, please contact the
> sender by
> reply email and destroy all copies of the original message.
>
> -----------------------------------------------------------------------------------
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Sep 19 2012 - 13:30:03 PDT
