Strong scaling performance, Collide benchmark, Haswell, 64M particles
Performance in millions of particle-timesteps / second / node
Nodes | CPU (mpi,hyper) | Kokkos/OMP (mpi,hyper,thread) | Kokkos/serial (mpi,hyper) | |
1 | 122.3 (64,2) | 123 (32,2,2) | 131.5 (64,2) | |
2 | 133.2 (64,2) | 135.2 (32,2,2) | 142.4 (64,2) | |
4 | 153 (64,2) | 150.9 (32,2,2) | 159.8 (64,2) | |
8 | 183.9 (64,2) | 174.3 (32,2,2) | 184.8 (64,2) | |
16 | 211.2 (64,2) | 201 (32,2,2) | 211.7 (64,2) | |
32 | 221.9 (64,2) | 211 (32,2,2) | 221.3 (64,2) | |
64 | 226.7 (64,2) | 220.3 (32,2,2) | 199.7 (32,1) |
1 | srun -n 64 -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.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=cpu.kind=strong.size=64M.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 256 -v y 160 -v z 160 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=cpu.kind=strong.size=64M.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.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=cpu.kind=strong.size=64M.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 160 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=cpu.kind=strong.size=64M.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 160 -v z 160 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=cpu.kind=strong.size=64M.node=16.mpi=64.hyper=2 |
32 | srun -n 2048 -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.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=cpu.kind=strong.size=64M.node=32.mpi=64.hyper=2 |
64 | srun -n 4096 -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.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=cpu.kind=strong.size=64M.node=64.mpi=64.hyper=2 |
1 | setenv OMP_NUM_THREADS 2; srun -n 32 -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.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=kokkos_omp.kind=strong.size=64M.node=1.mpi=32.thread=2.hyper=2 |
2 | setenv OMP_NUM_THREADS 2; srun -n 64 -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.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=kokkos_omp.kind=strong.size=64M.node=2.mpi=32.thread=2.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.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=kokkos_omp.kind=strong.size=64M.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 160 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=kokkos_omp.kind=strong.size=64M.node=8.mpi=32.thread=2.hyper=2 |
16 | setenv OMP_NUM_THREADS 2; srun -n 512 -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.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=kokkos_omp.kind=strong.size=64M.node=16.mpi=32.thread=2.hyper=2 |
32 | setenv OMP_NUM_THREADS 2; srun -n 1024 -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.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=kokkos_omp.kind=strong.size=64M.node=32.mpi=32.thread=2.hyper=2 |
64 | setenv OMP_NUM_THREADS 2; srun -n 2048 -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.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=kokkos_omp.kind=strong.size=64M.node=64.mpi=32.thread=2.hyper=2 |
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 256 -v y 160 -v z 160 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=kokkos_serial.kind=strong.size=64M.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 256 -v y 160 -v z 160 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=kokkos_serial.kind=strong.size=64M.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.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=kokkos_serial.kind=strong.size=64M.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 160 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=kokkos_serial.kind=strong.size=64M.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 160 -v z 160 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=kokkos_serial.kind=strong.size=64M.node=16.mpi=64.hyper=2 |
32 | srun -n 2048 -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.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=kokkos_serial.kind=strong.size=64M.node=32.mpi=64.hyper=2 |
64 | srun -n 2048 -C haswell --ntasks-per-node 32 --cpu_bind=rank -c 2 ./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.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=kokkos_serial.kind=strong.size=64M.node=64.mpi=32.hyper=1 |