Weak scaling performance, Sphere benchmark, Haswell, 16M particles/node
Performance in millions of particle-timesteps / second / node

Nodes CPU (mpi,hyper) Kokkos/OMP (mpi,hyper,thread) Kokkos/serial (mpi,hyper)
1 178.7 (64,2) 188.7 (16,2,4) 186.8 (64,2)
2 174.6 (64,2) 185.4 (16,2,4) 192.1 (64,2)
4 159 (64,2) 159.6 (32,2,2) 162.4 (64,2)
8 143.6 (64,2) 152.9 (32,2,2) 146.2 (64,2)
16 134.5 (64,2) 146.1 (16,2,4) 138 (64,2)

Run commands and logfile links for column CPU

1 srun -n 64 -C haswell --ntasks-per-node 64 --cpu_bind=rank -c 1 ./spa_mutrino_cpu -v x 128 -v y 80 -v z 160 -v t 100 -in in.sphere.steps -log log.sparta.date=23Dec17.model=sphere.machine=mutrino.pkg=cpu.kind=weak.size=16M.node=1.mpi=64.hyper=2
2 srun -n 128 -C haswell --ntasks-per-node 64 --cpu_bind=rank -c 1 ./spa_mutrino_cpu -v x 128 -v y 160 -v z 160 -v t 100 -in in.sphere.steps -log log.sparta.date=23Dec17.model=sphere.machine=mutrino.pkg=cpu.kind=weak.size=16M.node=2.mpi=64.hyper=2
4 srun -n 256 -C haswell --ntasks-per-node 64 --cpu_bind=rank -c 1 ./spa_mutrino_cpu -v x 256 -v y 160 -v z 160 -v t 100 -in in.sphere.steps -log log.sparta.date=23Dec17.model=sphere.machine=mutrino.pkg=cpu.kind=weak.size=16M.node=4.mpi=64.hyper=2
8 srun -n 512 -C haswell --ntasks-per-node 64 --cpu_bind=rank -c 1 ./spa_mutrino_cpu -v x 256 -v y 160 -v z 320 -v t 100 -in in.sphere.steps -log log.sparta.date=23Dec17.model=sphere.machine=mutrino.pkg=cpu.kind=weak.size=16M.node=8.mpi=64.hyper=2
16 srun -n 1024 -C haswell --ntasks-per-node 64 --cpu_bind=rank -c 1 ./spa_mutrino_cpu -v x 256 -v y 320 -v z 320 -v t 100 -in in.sphere.steps -log log.sparta.date=23Dec17.model=sphere.machine=mutrino.pkg=cpu.kind=weak.size=16M.node=16.mpi=64.hyper=2

Run commands and logfile links for column Kokkos/OMP

1 setenv OMP_NUM_THREADS 4; srun -n 16 -C haswell --ntasks-per-node 16 --cpu_bind=cores -c 4 ./spa_mutrino_kokkos_omp -sf kk -k on t 4 -pk kokkos reduction parallel/reduce comm classic -v x 128 -v y 80 -v z 160 -v t 100 -in in.sphere.steps -log log.sparta.date=23Dec17.model=sphere.machine=mutrino.pkg=kokkos_omp.kind=weak.size=16M.node=1.mpi=16.thread=4.hyper=2
2 setenv OMP_NUM_THREADS 4; srun -n 32 -C haswell --ntasks-per-node 16 --cpu_bind=cores -c 4 ./spa_mutrino_kokkos_omp -sf kk -k on t 4 -pk kokkos reduction parallel/reduce comm classic -v x 128 -v y 160 -v z 160 -v t 100 -in in.sphere.steps -log log.sparta.date=23Dec17.model=sphere.machine=mutrino.pkg=kokkos_omp.kind=weak.size=16M.node=2.mpi=16.thread=4.hyper=2
4 setenv OMP_NUM_THREADS 2; srun -n 128 -C haswell --ntasks-per-node 32 --cpu_bind=cores -c 2 ./spa_mutrino_kokkos_omp -sf kk -k on t 2 -pk kokkos reduction parallel/reduce comm classic -v x 256 -v y 160 -v z 160 -v t 100 -in in.sphere.steps -log log.sparta.date=23Dec17.model=sphere.machine=mutrino.pkg=kokkos_omp.kind=weak.size=16M.node=4.mpi=32.thread=2.hyper=2
8 setenv OMP_NUM_THREADS 2; srun -n 256 -C haswell --ntasks-per-node 32 --cpu_bind=cores -c 2 ./spa_mutrino_kokkos_omp -sf kk -k on t 2 -pk kokkos reduction parallel/reduce comm classic -v x 256 -v y 160 -v z 320 -v t 100 -in in.sphere.steps -log log.sparta.date=23Dec17.model=sphere.machine=mutrino.pkg=kokkos_omp.kind=weak.size=16M.node=8.mpi=32.thread=2.hyper=2
16 setenv OMP_NUM_THREADS 4; srun -n 256 -C haswell --ntasks-per-node 16 --cpu_bind=cores -c 4 ./spa_mutrino_kokkos_omp -sf kk -k on t 4 -pk kokkos reduction parallel/reduce comm classic -v x 256 -v y 320 -v z 320 -v t 100 -in in.sphere.steps -log log.sparta.date=23Dec17.model=sphere.machine=mutrino.pkg=kokkos_omp.kind=weak.size=16M.node=16.mpi=16.thread=4.hyper=2

Run commands and logfile links for column Kokkos/serial

1 srun -n 64 -C haswell --ntasks-per-node 64 --cpu_bind=rank -c 1 ./spa_mutrino_kokkos_serial -sf kk -k on -pk kokkos reduction parallel/reduce comm classic -v x 128 -v y 80 -v z 160 -v t 100 -in in.sphere.steps -log log.sparta.date=23Dec17.model=sphere.machine=mutrino.pkg=kokkos_serial.kind=weak.size=16M.node=1.mpi=64.hyper=2
2 srun -n 128 -C haswell --ntasks-per-node 64 --cpu_bind=rank -c 1 ./spa_mutrino_kokkos_serial -sf kk -k on -pk kokkos reduction parallel/reduce comm classic -v x 128 -v y 160 -v z 160 -v t 100 -in in.sphere.steps -log log.sparta.date=23Dec17.model=sphere.machine=mutrino.pkg=kokkos_serial.kind=weak.size=16M.node=2.mpi=64.hyper=2
4 srun -n 256 -C haswell --ntasks-per-node 64 --cpu_bind=rank -c 1 ./spa_mutrino_kokkos_serial -sf kk -k on -pk kokkos reduction parallel/reduce comm classic -v x 256 -v y 160 -v z 160 -v t 100 -in in.sphere.steps -log log.sparta.date=23Dec17.model=sphere.machine=mutrino.pkg=kokkos_serial.kind=weak.size=16M.node=4.mpi=64.hyper=2
8 srun -n 512 -C haswell --ntasks-per-node 64 --cpu_bind=rank -c 1 ./spa_mutrino_kokkos_serial -sf kk -k on -pk kokkos reduction parallel/reduce comm classic -v x 256 -v y 160 -v z 320 -v t 100 -in in.sphere.steps -log log.sparta.date=23Dec17.model=sphere.machine=mutrino.pkg=kokkos_serial.kind=weak.size=16M.node=8.mpi=64.hyper=2
16 srun -n 1024 -C haswell --ntasks-per-node 64 --cpu_bind=rank -c 1 ./spa_mutrino_kokkos_serial -sf kk -k on -pk kokkos reduction parallel/reduce comm classic -v x 256 -v y 320 -v z 320 -v t 100 -in in.sphere.steps -log log.sparta.date=23Dec17.model=sphere.machine=mutrino.pkg=kokkos_serial.kind=weak.size=16M.node=16.mpi=64.hyper=2