Strong scaling performance, Collide benchmark, 8M particles
Performance in millions of particle-timesteps / second / node
Nodes | SandyBridge | Haswell | Broadwell | KNL | K80-1 | P100-1 | |
1 | 80 (Kokkos/serial,mpi=16) | 191.8 (Kokkos/serial,mpi=64,hyper=2) | 206.7 (Kokkos/serial,mpi=72,hyper=2) | 294.8 (Kokkos/KNL,mpi=128,thread=2,hyper=4) | 145.6 (Kokkos/Cuda,mpi=2) | 301.2 (Kokkos/Cuda,mpi=1) | |
2 | 94.1 (Kokkos/serial,mpi=16) | 221.6 (Kokkos/serial,mpi=64,hyper=2) | 239.3 (CPU,mpi=72,hyper=2) | 314.9 (Kokkos/KNL,mpi=64,thread=4,hyper=4) | 147.1 (Kokkos/Cuda,mpi=2) | 277.4 (Kokkos/Cuda,mpi=1) | |
4 | 115.8 (Kokkos/serial,mpi=16) | 235.7 (Kokkos/serial,mpi=64,hyper=2) | 254.2 (CPU,mpi=72,hyper=2) | 309.7 (Kokkos/KNL,mpi=64,thread=4,hyper=4) | 120.5 (Kokkos/Cuda,mpi=2) | 241.4 (Kokkos/Cuda,mpi=1) | |
8 | 131.2 (Kokkos/serial,mpi=16) | 263.2 (CPU,mpi=64,hyper=2) | 297 (CPU,mpi=72,hyper=2) | 261.1 (Kokkos/KNL,mpi=32,thread=8,hyper=4) | 103.2 (Kokkos/Cuda,mpi=2) | 194.5 (Kokkos/Cuda,mpi=1) | |
16 | 134.9 (Kokkos/serial,mpi=16) | 439 (CPU,mpi=64,hyper=2) | 534.5 (CPU,mpi=72,hyper=2) | 212.7 (Kokkos/KNL,mpi=64,thread=4,hyper=4) | None | None | |
32 | 202.8 (CPU,mpi=16) | 399.9 (CPU,mpi=64,hyper=2) | 435.5 (CPU,mpi=72,hyper=2) | 162.3 (CPU/KNL,mpi=64,hyper=1) | None | None | |
64 | 191.5 (CPU,mpi=16) | 313.3 (CPU,mpi=32,hyper=1) | 353.7 (CPU,mpi=36,hyper=1) | 125.9 (CPU/KNL,mpi=64,hyper=1) | None | None |
1 | mpirun -n 16 -N 16 --bind-to core spa_chama_kokkos_serial -sf kk -k on -pk kokkos reduction parallel/reduce comm classic -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=chama.pkg=kokkos_serial.kind=strong.size=8M.node=1.mpi=16 |
2 | mpirun -n 32 -N 16 --bind-to core spa_chama_kokkos_serial -sf kk -k on -pk kokkos reduction parallel/reduce comm classic -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=chama.pkg=kokkos_serial.kind=strong.size=8M.node=2.mpi=16 |
4 | mpirun -n 64 -N 16 --bind-to core spa_chama_kokkos_serial -sf kk -k on -pk kokkos reduction parallel/reduce comm classic -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=chama.pkg=kokkos_serial.kind=strong.size=8M.node=4.mpi=16 |
8 | mpirun -n 128 -N 16 --bind-to core spa_chama_kokkos_serial -sf kk -k on -pk kokkos reduction parallel/reduce comm classic -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=chama.pkg=kokkos_serial.kind=strong.size=8M.node=8.mpi=16 |
16 | mpirun -n 256 -N 16 --bind-to core spa_chama_kokkos_serial -sf kk -k on -pk kokkos reduction parallel/reduce comm classic -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=chama.pkg=kokkos_serial.kind=strong.size=8M.node=16.mpi=16 |
32 | mpirun -n 512 -N 16 --bind-to core spa_chama_cpu -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=chama.pkg=cpu.kind=strong.size=8M.node=32.mpi=16 |
64 | mpirun -n 1024 -N 16 --bind-to core spa_chama_cpu -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=chama.pkg=cpu.kind=strong.size=8M.node=64.mpi=16 |
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 80 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=kokkos_serial.kind=strong.size=8M.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 80 -v z 80 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=kokkos_serial.kind=strong.size=8M.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 128 -v y 80 -v z 80 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=kokkos_serial.kind=strong.size=8M.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 128 -v y 80 -v z 80 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=cpu.kind=strong.size=8M.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 128 -v y 80 -v z 80 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=cpu.kind=strong.size=8M.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 128 -v y 80 -v z 80 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=cpu.kind=strong.size=8M.node=32.mpi=64.hyper=2 |
64 | srun -n 2048 -C haswell --ntasks-per-node 32 --cpu_bind=rank -c 2 ./spa_mutrino_cpu -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=cpu.kind=strong.size=8M.node=64.mpi=32.hyper=1 |
1 | mpiexec -np 72 -npernode 72 --oversubscribe --bind-to core ./spa_serrano_kokkos_serial -sf kk -k on -pk kokkos reduction parallel/reduce comm classic -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=serrano.pkg=kokkos_serial.kind=strong.size=8M.node=1.mpi=72.hyper=2 |
2 | mpiexec -np 144 -npernode 72 --oversubscribe --bind-to core ./spa_serrano_cpu -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=serrano.pkg=cpu.kind=strong.size=8M.node=2.mpi=72.hyper=2 |
4 | mpiexec -np 288 -npernode 72 --oversubscribe --bind-to core ./spa_serrano_cpu -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=serrano.pkg=cpu.kind=strong.size=8M.node=4.mpi=72.hyper=2 |
8 | mpiexec -np 576 -npernode 72 --oversubscribe --bind-to core ./spa_serrano_cpu -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=serrano.pkg=cpu.kind=strong.size=8M.node=8.mpi=72.hyper=2 |
16 | mpiexec -np 1152 -npernode 72 --oversubscribe --bind-to core ./spa_serrano_cpu -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=serrano.pkg=cpu.kind=strong.size=8M.node=16.mpi=72.hyper=2 |
32 | mpiexec -np 2304 -npernode 72 --oversubscribe --bind-to core ./spa_serrano_cpu -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=serrano.pkg=cpu.kind=strong.size=8M.node=32.mpi=72.hyper=2 |
64 | mpiexec -np 2304 -npernode 36 --oversubscribe --bind-to core ./spa_serrano_cpu -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=serrano.pkg=cpu.kind=strong.size=8M.node=64.mpi=36.hyper=1 |
1 | setenv OMP_NUM_THREADS 2; srun -n 128 -C knl --ntasks-per-node 128 --cpu_bind=threads -c 2 ./spa_mutrino_kokkos_knl -sf kk -k on t 2 -pk kokkos reduction parallel/reduce comm classic -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=kokkos_knl.kind=strong.size=8M.node=1.mpi=128.thread=2.hyper=4 |
2 | setenv OMP_NUM_THREADS 4; srun -n 128 -C knl --ntasks-per-node 64 --cpu_bind=cores -c 4 ./spa_mutrino_kokkos_knl -sf kk -k on t 4 -pk kokkos reduction parallel/reduce comm classic -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=kokkos_knl.kind=strong.size=8M.node=2.mpi=64.thread=4.hyper=4 |
4 | setenv OMP_NUM_THREADS 4; srun -n 256 -C knl --ntasks-per-node 64 --cpu_bind=cores -c 4 ./spa_mutrino_kokkos_knl -sf kk -k on t 4 -pk kokkos reduction parallel/reduce comm classic -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=kokkos_knl.kind=strong.size=8M.node=4.mpi=64.thread=4.hyper=4 |
8 | setenv OMP_NUM_THREADS 8; srun -n 256 -C knl --ntasks-per-node 32 --cpu_bind=cores -c 8 ./spa_mutrino_kokkos_knl -sf kk -k on t 8 -pk kokkos reduction parallel/reduce comm classic -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=kokkos_knl.kind=strong.size=8M.node=8.mpi=32.thread=8.hyper=4 |
16 | setenv OMP_NUM_THREADS 4; srun -n 1024 -C knl --ntasks-per-node 64 --cpu_bind=cores -c 4 ./spa_mutrino_kokkos_knl -sf kk -k on t 4 -pk kokkos reduction parallel/reduce comm classic -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=kokkos_knl.kind=strong.size=8M.node=16.mpi=64.thread=4.hyper=4 |
32 | srun -n 2048 -C knl --ntasks-per-node 64 --cpu_bind=rank -c 4 ./spa_mutrino_knl -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=cpu_knl.kind=strong.size=8M.node=32.mpi=64.hyper=1 |
64 | srun -n 4096 -C knl --ntasks-per-node 64 --cpu_bind=rank -c 4 ./spa_mutrino_knl -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.steps -log log.sparta.date=23Dec17.model=collide.machine=mutrino.pkg=cpu_knl.kind=strong.size=8M.node=64.mpi=64.hyper=1 |
1 | mpirun -np 2 --npersocket 1 --bind-to core spa_ride80_kokkos_cuda -sf kk -k on g 2 -pk kokkos reduction atomic comm threaded -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.gpu.steps -log log.sparta.date=23Dec17.model=collide.machine=ride80.pkg=kokkos_cuda.kind=strong.size=8M.node=1.mpi=2.gpu=2 |
2 | mpirun -np 4 --npersocket 1 --bind-to core spa_ride80_kokkos_cuda -sf kk -k on g 2 -pk kokkos reduction atomic comm threaded -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.gpu.steps -log log.sparta.date=23Dec17.model=collide.machine=ride80.pkg=kokkos_cuda.kind=strong.size=8M.node=2.mpi=2.gpu=2 |
4 | mpirun -np 8 --npersocket 1 --bind-to core spa_ride80_kokkos_cuda -sf kk -k on g 2 -pk kokkos reduction atomic comm threaded -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.gpu.steps -log log.sparta.date=23Dec17.model=collide.machine=ride80.pkg=kokkos_cuda.kind=strong.size=8M.node=4.mpi=2.gpu=2 |
8 | mpirun -np 16 --npersocket 1 --bind-to core spa_ride80_kokkos_cuda -sf kk -k on g 2 -pk kokkos reduction atomic comm threaded -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.gpu.steps -log log.sparta.date=23Dec17.model=collide.machine=ride80.pkg=kokkos_cuda.kind=strong.size=8M.node=8.mpi=2.gpu=2 |
16 | None |
32 | None |
64 | None |
1 | mpirun -np 1 --npernode 1 --bind-to core spa_ride100_kokkos_cuda -sf kk -k on g 1 -pk kokkos reduction atomic comm threaded -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.gpu.steps -log log.sparta.date=23Dec17.model=collide.machine=ride100.pkg=kokkos_cuda.kind=strong.size=8M.node=1.mpi=1.gpu=1 |
2 | mpirun -np 2 --npernode 1 --bind-to core spa_ride100_kokkos_cuda -sf kk -k on g 1 -pk kokkos reduction atomic comm threaded -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.gpu.steps -log log.sparta.date=23Dec17.model=collide.machine=ride100.pkg=kokkos_cuda.kind=strong.size=8M.node=2.mpi=1.gpu=1 |
4 | mpirun -np 4 --npernode 1 --bind-to core spa_ride100_kokkos_cuda -sf kk -k on g 1 -pk kokkos reduction atomic comm threaded -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.gpu.steps -log log.sparta.date=23Dec17.model=collide.machine=ride100.pkg=kokkos_cuda.kind=strong.size=8M.node=4.mpi=1.gpu=1 |
8 | mpirun -np 8 --npernode 1 --bind-to core spa_ride100_kokkos_cuda -sf kk -k on g 1 -pk kokkos reduction atomic comm threaded -v x 128 -v y 80 -v z 80 -v t 100 -in in.collide.gpu.steps -log log.sparta.date=23Dec17.model=collide.machine=ride100.pkg=kokkos_cuda.kind=strong.size=8M.node=8.mpi=1.gpu=1 |
16 | None |
32 | None |
64 | None |