[Wrf-users] Optimizing OMP_NUM_THREADS

Spencer Swift swift at sgi.com
Fri Jul 24 15:08:25 MDT 2015


Brian,

>> Options were for dm+sm (option 55 pgf90/pgcc) and basic nesting (option 1)

Also, since you built for both MPI (DM) and OMP (SM) parallelism, the MPI scaling is typically much better than the OMP scaling. And when the MPI scaling starts to taper off, using hybrid MPI tasks with OMP threads can often extend the scaling further.

Spencer Swift



More information about the Wrf-users mailing list