Basic sample submission scripts are listed HERE.
Note that you also can reserve the nodes with command salloc
salloc -N1 -n4 -t 12:00:00 — request 4 CPU-cores (n4) on one node (N1) for 12 hours in default queue (“all”)
salloc -p long -n24 -t 48:00:00 — request 24 CPU-cores for 48 hours in queue “long” (on whatever number of nodes)
salloc -p NGPU --gres=gpu:2 -N1 -n2 -t 12:00:00 — request 2 CPU cores and two GPUs in queue NGPU (we have 2 GPUs/node max) for 12 hours
salloc -p Broadwell -N1 -w zeus300 --exclusive — request particular node (zeus300) in exclusive mode (all resources)
When the node is reserved, you can send e.g. mpi code with mpirun command, or just login to the target node with ssh and work interactively in the command line.
You can also use srun command to run tasks on CPUs, reserved with salloc :
[aa3025@zeus2 ~]$ qstat
JOBID PARTITION NAME USER ST TIME NODES NODELIST(REASON)
538907 NGPU bash aa3025 R 0:04 1 zeus405
[aa3025@zeus2 ~]$ srun -n 4 hostname
zeus405
zeus405
zeus405
zeus405
0 Comments.