Since a non-docker setup doesn't have the cgroups limiting capabilities, it would be nice to have finer control over the relaxation step performed with OpenMM.
The default AF script uses OpenMM with a CPU platform which by default consumes all CPUs. The behavior can be tuned with the OPENMM_CPU_THREADS environment variable which can be setup similar to the GPU selection process.
For our local version I added following lines:
c)
openmm_threads=$OPTARG
;;
<...>
if [[ "$openmm_threads" ]] ; then
export OPENMM_CPU_THREADS=${openmm_threads}
fi
Since a non-docker setup doesn't have the cgroups limiting capabilities, it would be nice to have finer control over the relaxation step performed with OpenMM.
The default AF script uses OpenMM with a CPU platform which by default consumes all CPUs. The behavior can be tuned with the
OPENMM_CPU_THREADS
environment variable which can be setup similar to the GPU selection process.For our local version I added following lines: