[slurm-dev] “srun -n“ allocate same gpus to multi tasks

[Qianqian Sha]

Hi:

    I'm a new user to slurm and I'm interested in gpu scheduling.

    When I run command "srun -n 2 --gres=gpu:2 ./test.sh"

        test.sh:
        echo $CUDA_VISIBLE_DEVICES

     we'll see
        0,1
        0,1

     These sub-jobs use the same gpu. But I think the result should be
        0,1
        2,3

     Is that right?

Thanks.