Mdrun use 1MPI thread and 8 OMP threads and my system has 1 socket 4 cores and 8 logical processors
I think 1 MPI is 1 core but i have 4 so is there any way i can utilise them all and boost my mdrun so that it takes less time
I think 1 MPI is 1 core but i have 4 so is there any way i can utilise them all and boost my mdrun so that it takes less time
Hi!
If you have a few CPU cores you can usually get better performance by focusing on domain decomposition (through thread-MPI) rather than regular Open-MP parallelization, particularly if you have a larger simulation system. You can increase the number thread-MPI threads through the -ntmpi flag, and it might also be beneficial to turn on thread pinning (set -pin on). Which settings are best always depend on your simulation system and your hardware setup, so it’s always a good idea to do a small benchmark and test a few different configurations before you launch longer runs. You can read more about setting that can impact performance here: https://manual.gromacs.org/documentation/current/user-guide/mdrun-performance.html
Hope this helped!