Hello, I am installing the slurm-17.02.6 on my testcluster, and i will run my software on this cluster. On my cluster, each computing node have been configured to have 4 GPUs.
Now i meet three problem as below: 1. How can i specify the different GPUs request on different nodes in slurm scripts, which will be submitted by command sbatch? I have found the argument "--gres", but the specified resources will be allocated to the job on each node. for example , i want to run my jobs on 2 nodes and 7 GPUs (4 GPUs on node1, and 3 GPUs on node2) . Can slurm 17.02.6 support this ? Or Can i specified the total GPU request for the whole job (not each node)? 2. Can slurm support to specify (or configure) the rate of cpu/gpu in the same job? For example, While 1 GPU is allocated to job, 2 CPU on the same node will be allocated to it. 3. Similar to question 1, Can i specify different CPUs request on different node? For example, i want to use 20 CPU(of 2 nodes) to run my job, 12 CPUs on node1 and 8 CPUs on node2 Sorry to have asked so many questions. best regards