【How to use GPU to accelerate PME calculation?】

User discussions
, , ,
1

GROMACS version: 2021.4
GROMACS modification: No

Hi, I am new to gromacs and recently want to try out Martini3 force field.
Therefore, I perform some tutorials on the Martini and gromacs website.
I found that running on CPU takes times. The GPU on my computer is nvidia 1660ti which can provide better performance than my 4800h CPU. I also have my server running RTX3070 and AMD 5950X.
I use the command:
gmx mdrun -v -nt 0 -nb gpu -pme gpu -bonded gpu -update gpu -deffnm 3-run
Try to perform all the calculation on the GPU, but it did not work…
The error is: Feature not implemented:
Cannot compute PME interactions on a GPU, because PME GPU does not support
systems that do not use PME for electrostatics.

But if I switch back to:
gmx mdrun -v -nt 0 -nb gpu -pme cpu -deffnm 3-run
The calculation can be done on both CPU and GPU.

My question is how to make the best use of both hardware?
How can I decide how to split the work between GPU and CPU? And, what will be the command to do so?

Additionally, could you suggest a tutorial for me to learn on the format of .top, .itp, .mdp and the md simulation and analysis with gromacs?
Since I am new to MD, I would like to learn and understand the meaning of parameters set in the input files.
Thank you very much for your great work and support!

Regards,
SWL

2

Unless you use PME for electrostatics, there isn’t much else you can do to speed things up. Presumably you’re using reaction-field for long-range electrostatics, which may be required by MARTINI but offhand I’m not a CG person so I can’t say for sure.

Lots of them at GROMACS Tutorials (the first one covers the super basics like this), as well as From Proteins to Perturbed Hamiltonians: A Suite of Tutorials for the GROMACS-2018 Molecular Simulation Package [Article v1.0] | Living Journal of Computational Molecular Science

3

Thank you very much for your reply.
I will try to use PME in mdp file and I will also consult with Martini community of how to speed up the performance.
Thank you for your suggestion on the tutorial. I will check them out.