On Pinnacle/Karpinski we currently only support interactive jobs on single nodes from one core to the number of the cores in the node. So always –nodes=1
with srun
. MPI is supported, just not multiple-node interactive MPI.
srun --nodes=1 --ntasks-per-node=1 --cpus-per-task=32 --partition comp06 --time=6:00:00 --pty /bin/bash