[HTCondor-users] Using HTCondor running mpirun

[JIANG Xiaowei <jiangxw@xxxxxxxxxx>]

Dear ALLï

We encountered a problem when using GPU node to run mpirun. If several mpirun jobs submitted and run on a same worker node, only the first CPUs are used.

The parameter in mpirun called '--bind-to none' can avoid this. But I would still like to know if HTCondor has a way to allocate different CPUs to each job (compared with SLURM cluster we are running meantime, SLURM can keep different job using different CPUs.)

Any hint or suggestion would be appreciated!

Thanks,

Xiaowei