Intel mpirun interpreting process count incorrectly in Slurm

[Martijn_D_]

mpirun launches only a maximum 8 processes per node, on a node that has 12 cores.

I suspect this is a result of misinterpretation of the processor count from the slurm environment

or some kind of limitation enforced by mpirun.

Intel MPI version:4.0.3.008

Slurm version : 2.2.4

Nodes have Intel Xeon CPU X5650 @ 2.67Ghz (12 cores)

1. Request resources via slurm.

salloc -n 24

The resources are allocated properly.

echo $SLURM_NNODES

2

echo $SLURM_JOB_NUM_NODES

2

echo $SLURM_NPROCS

24

echo $SLURM_TASKS_PER_NODE

12(x2)

echo #SLURM_JOB_NODELIST

node[049-050]

2. Run mpi job

mpirun ./intelmpijob

This lauches 8 processes pernode (total of 16).

Any quick response will be greatly appreciated.

[Dmitry_S_Intel]

Dear martijn@clustervision.com,

Thank you forthe post.

Please try to run your job with

mpiexec.hydra -bootstrap=slurm ./intelmpijob

Best regards,
Dmitry