For large systems of over 80 CPUs there is a large amount of memory contention in gomp_team_barrier_wait_end that impacts on performance and unnecessarily increases run time. This can be resolved by setting GOMP_SPINCOUNT to 7000 to reduce the contention. Also ensure that the OMP_WAIT_POLICY is set to active to ensure the optimial OMP wait policy is used for larger systems.
For large systems of over 80 CPUs there is a large amount of memory contention in gomp_team_barrier_wait_end that impacts on performance and unnecessarily increases run time. This can be resolved by setting GOMP_SPINCOUNT to 7000 to reduce the contention. Also ensure that the OMP_WAIT_POLICY is set to active to ensure the optimial OMP wait policy is used for larger systems.