[Wrf-users] Openmpi or mich2

[Craig Mattocks]

On Sat, Sep 29, 2012 at 7:02 AM, ugo merlini <ugomerlini at hotmail.com> wrote:

> which is the best for a single computer (core i7-3770) and wrf 3.4.1?

In my experience on multiple systems, from large supercomputing clusters to laptops, mpich2 runs a lot faster.

Performance (seconds) of a 12-hour WRF 3.4.1 simulation (one grid domain) on a MacBook Pro laptop (Intel core i7, 4 cores, Mac OS X 10.8.2 Mountain Lion) Intel compilers, mpich2:

1. setenv OMP_NUM_THREADS 1
/usr/bin/time mpirun -np 4 ./wrf.exe >& wrf.out &
      30.16 real       116.74 user         3.50 sys

2. setenv OMP_NUM_THREADS 2
/usr/bin/time mpirun -np 4 ./wrf.exe >& wrf.out &
      28.99 real       219.63 user         4.46 sys

3. setenv OMP_NUM_THREADS 2
/usr/bin/time mpirun -np 2 ./wrf.exe >& wrf.out &
      24.35 real        94.33 user         1.12 sys

4. setenv OMP_NUM_THREADS 4
/usr/bin/time mpirun -np 1 ./wrf.exe >& wrf.out &
      23.94 real        93.13 user         0.47 sys

5. setenv OMP_NUM_THREADS 4
/usr/bin/time mpirun -np 2 ./wrf.exe > & wrf.out &
      23.16 real       176.09 user         1.71 sys

Winner seems to be pure OpenMP (run #4) on this platform.

Best,
Craig