I just got something similar working. I used CR_CPU for SelectTypeParameters.

Marc Rollins <[email protected]> writes:

> Hello,
>
> I am attempting to run multiple jobs simultaneously on a node with two GPUs. 
> However, all my attempts fail. Both
> jobs are queued, but only one runs at a time while the other remains in the 
> queue. My SLURM configuration is
> below. Any assistance is greatly appreciated.
>
> # slurm.conf
> GresTypes=gpu,tfcpu
>
> SelectType=select/cons_res
> SelectTypeParameters=CR_Core_Memory
>
> NodeName=Hulk NodeAddr=10.0.1.10 Gres=gpu:2,tfcpu:1 CPUs=12 SocketsPerBoard=1 
> CoresPerSocket=6 ThreadsPerCore=2
> RealMemory=15886 State=UNKNOWN Weight=1
>
> # gres.conf
> Name=gpu File=/dev/nvidia0
> Name=gpu File=/dev/nvidia1
> Name=tfcpu
>
> # invocation
> sbatch --gres=gpu:1 myjob.bash
> sbatch --gres=gpu:1 myjob.bash
>
> --
> Marc Rollins                                                            
> Software Engineer                                                       
> o: (509) 863-9228                                                       
> [gj_email_s]                                                            
> Entrepreneur Magazine's 'Best Entrepreneurial Companies in America 2015'
>

Reply via email to