Entering edit mode
3.5 years ago
Juke34
8.9k
Using parallelization in a proper way is not always straightforward. Here some help to better understand how it works with Slurm:
(Full part of this documentation is available here)
If the cluster has two type of nodes, e.g. 16 and 36 CPUs nodes. When asking for 36 CPU for a single job you must use the following syntax otherwise the job may be sent to a 16 CPU node:
srun -J jobName [-N 1] -n 1 -c 36 <command>
-N
is optional because -c 36
means you need 36 cores in one node, since threads cannot span across nodes. But it doesn't hurt to be explicit and specify all three parameters.