Slow molecular dynamics simulations of protein in water with GPU

GROMACS version:
GROMACS modification: Yes/No
Here post your question

I run GROMACS 2023.2 GPU built on HPC using Nvidia hpcx package. I have successfully simulated protein ligand simulations for 400 ns. Right now, i am trying to simulate same protein in water for apo protein comparison using same script file, it is taking abnormal time to finish it. I don’t know why it’s taking too much time.
Command Line: mpirun -np xx gmx_mpi mdrun --deffnm md_0_10 -cpi md_0_10.cpt -noappend -v -nb gpu
Should i exclude non-bonded interactions with gpu by removing flag -nb gpu at the end of the gmx_mpi command?
Please suggest me the correction. Thank you in advance.


Sorry for the late answer. What do you mean by it’s taking an abnormal time to finish the calculations? It’s difficult to say exactly why your calculations are slow without more details or benchmarks on your specific HPC system. Can you share any performance-related messages that should be printed in the log file? Then, it’s probably a good idea to try to run without the GPU and see if you get better performance. How many MPI ranks and threads do you use? Perhaps your selection here is non-optimal so you could try to vary these values (if your system is large it’s usually a good idea to use more MPI ranks than OpenMP threads on each node). Since you’re running on a HPC system it might be a good idea to contact any support staff there in case something is wrong with how GROMACS was built on that particular system.

See if it helps with one or two ranks per GPU, presumably mpirun -np 1 ....