GROMACS version: 2022.3
GROMACS modification: No
Here post your question
If I do a hybrid MPI + OpenMP build and set OMP_NUM_THREADS=1, will there be a significant performance difference with a version built purely with MPI only? Even if I only spawn one OpenMP thread, there will still be OpenMP overheads. I know I can run a simple benchmark, but the significant performance difference may only show on certain tests. Has anyone benchmarked these two different versions?
Regards,