Translations:GROMACS/35/en

• GROMACS versions 2020.0 up to and including 2021.5 contain a bug when used on GPUs of Volta or newer generations (i.e. V100, T4 and A100) with mdrun option -update gpu that could have perturbed the virial calculation and, in turn, led to incorrect pressure coupling. The GROMACS developers state in the 2021.6 Release Notes:

  The GPU update is not enabled by default, so the error can only appear in simulations where it was manually selected, and even in this case the error might be rare since we have not observed it in practice in the testing we have performed.

  Further discussion of this bug can be found in the GitLab issue #4393 of the GROMACS project.
• Our clusters (Beluga, Cedar, Graham and Narval) have differently configured GPU nodes.
  On the page Using GPUs with Slurm#Available GPUs you can find more information about the different node configurations (GPU models and number of GPUs and CPUs per node).
• GROMACS imposes a number of constraints for choosing number of GPUs, tasks (MPI ranks) and OpenMP threads.
  For GROMACS 2018.2 the constraints are:

• The number of --tasks-per-node always needs to be the same as, or a multiple of the number of GPUs (--gpus-per-node).
• GROMACS will not run GPU runs with only 1 OpenMP thread, unless forced by setting the -ntomp option.
  According to GROMACS developers, the optimum number of --cpus-per-task is between 2 and 6.

• Avoid using a larger fraction of CPUs and memory than the fraction of GPUs you have requested in a node.

Message Passing Interface