Hello Gromacs users,
I have a general question on how to implement alchemical calculations on a GPU cluster. Our GPU cluster has 8 GPUs per node (A100) with 128 cores (2 nodes of AMD epyc). I want to run 8 replicas at different temperatures for one window in the alchemical transformation. Here is my command to run the simulations:
Inconsistency in user input: There were 8 GPU tasks assigned on node gpu1, but 16 GPU tasks were identified, and these must match. Reconsider your GPU task assignment, number of ranks, or your use of the -nb, -pme, and -npme options.
All I want is to partition the 8 replicas so that each replica runs with 16 cores of CPU and 1 core of GPU. Can I get some suggestions from the Gromacs community?
The manual is incorrect here. You also need to specify the PME tasks, so 0011223344556677. But I think it should also work without specifying the -gputasks option.
Thanks for the information. Then I got the error message like this:
Fatal error:
The -dd or -npme option request a parallel simulation, but gmx mdrun was not started through mpirun/mpiexec or only one rank was requested through mpirun/mpiexec