GROMACS version: 2020.5
GROMACS modification: No
Hi everyone, I have a question regarding the GPU/PBS usage with gromacs.
I am running my simulations in one node that has four GPUs and 80 cores. General, I am running four simulations at the same time in this node, using 16 cores for each simulation (8 mpi/2 openMP). So, my PBS script file has:
#PBS -l select=1:ncpus=16:mpiprocs=8:ompthreads=2:ngpus=1
But in this case, all the simulations are using all the four GPUs at the same time.
My question is the following; is possible to control by the PBS file the number of GPUs Gromacs is going to use? I know that I can add the mdrun flag “gpu_id XX”. But this node is part of a big HPC server, so, I would like to control the number of GPUs by the PBS submission.
Thank you