Skip to content

Guide for NAMD3 Application

Select NAMD3 Docker Application.

Navigate to the Application Configuration screen.

Copy the following code block in the Run Script text box to run apoa1 benchmark concurrently on all GPUs on an 8-GPU compute node.

apoa1 benchmark

cd /examples
BENCHMARK=/examples/apoa1/apoa1.namd
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 0 --CUDASOAintegrate on +devices 0 2>&1 > gpu0 &
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 8 --CUDASOAintegrate on +devices 1 2>&1 > gpu1 &
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 16 --CUDASOAintegrate on +devices 2 2>&1 > gpu2 &
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 32 --CUDASOAintegrate on +devices 3 2>&1 > gpu3 &
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 64 --CUDASOAintegrate on +devices 4 2>&1 > gpu4 &
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 72 --CUDASOAintegrate on +devices 5 2>&1 > gpu5 &
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 80 --CUDASOAintegrate on +devices 6 2>&1 > gpu6 &
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 88 --CUDASOAintegrate on +devices 7 2>&1 > gpu7
wait
./ns_per_day.py gpu0
./ns_per_day.py gpu1
./ns_per_day.py gpu2
./ns_per_day.py gpu3
./ns_per_day.py gpu4
./ns_per_day.py gpu5
./ns_per_day.py gpu6
./ns_per_day.py gpu7

1.How to compute Figure of Merit (POM)

From the completed Workloads list, select your completed NAMD3 job. Select the STDOUT tab to view stdout logs. The FOM for NAMD3 8-GPU workload is the sum of the eight Nanoseconds per day numbers. Here is an example output from apoa1 benchmark run. The computed FOM for this example is 862.694 Nanoseconds per day (8 * 107.9 approx.).


Nanoseconds per day: 107.972
Mean time per step: 0.00160041
Standard deviation: 1.9379e-05
Nanoseconds per day: 107.949
Mean time per step: 0.00160076
Standard deviation: 1.59461e-05
Nanoseconds per day: 107.228
Mean time per step: 0.00161151
Standard deviation: 1.59286e-05
Nanoseconds per day: 107.572
Mean time per step: 0.00160637
Standard deviation: 6.90183e-06
Nanoseconds per day: 107.163
Mean time per step: 0.00161249
Standard deviation: 1.22992e-05
Nanoseconds per day: 108.366
Mean time per step: 0.0015946
Standard deviation: 1.39513e-05
Nanoseconds per day: 108.35
Mean time per step: 0.00159483
Standard deviation: 1.61872e-05
Nanoseconds per day: 108.094
Mean time per step: 0.00159861
Standard deviation: 1.68244e-05

jac benchmark

Copy the following code block in the Run Script text to run jac benchmark concurrently on all GPUs on an 8-GPU compute node.

cd /examples
BENCHMARK=/examples/jac/jac.namd
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 0 --CUDASOAintegrate on +devices 0 2>&1 > gpu0 &
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 8 --CUDASOAintegrate on +devices 1 2>&1 > gpu1 &
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 16 --CUDASOAintegrate on +devices 2 2>&1 > gpu2 &
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 32 --CUDASOAintegrate on +devices 3 2>&1 > gpu3 &
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 64 --CUDASOAintegrate on +devices 4 2>&1 > gpu4 &
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 72 --CUDASOAintegrate on +devices 5 2>&1 > gpu5 &
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 80 --CUDASOAintegrate on +devices 6 2>&1 > gpu6 &
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 88 --CUDASOAintegrate on +devices 7 2>&1 > gpu7
wait
./ns_per_day.py gpu0
./ns_per_day.py gpu1
./ns_per_day.py gpu2
./ns_per_day.py gpu3
./ns_per_day.py gpu4
./ns_per_day.py gpu5
./ns_per_day.py gpu6
./ns_per_day.py gpu7

2.How to compute Figure of Merit (FOM)

From the completed Workloads list, select your completed NAMD3 job. Select the STDOUT tab to view stdout logs. The FOM for NAMD3 8-GPU workload is the sum of the eight Nanoseconds per day numbers.

f1atpase benchmark

Copy the following code block in the Run Script text box to run f1atpase benchmark concurrently on all GPUs on an 8-GPU compute node.

cd /examples
BENCHMARK=/examples/f1atpase/f1atpase.namd
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 0 --CUDASOAintegrate on +devices 0 2>&1 > gpu0 &
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 8 --CUDASOAintegrate on +devices 1 2>&1 > gpu1 &
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 16 --CUDASOAintegrate on +devices 2 2>&1 > gpu2 &
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 32 --CUDASOAintegrate on +devices 3 2>&1 > gpu3 &
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 64 --CUDASOAintegrate on +devices 4 2>&1 > gpu4 &
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 72 --CUDASOAintegrate on +devices 5 2>&1 > gpu5 &
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 80 --CUDASOAintegrate on +devices 6 2>&1 > gpu6 &
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 88 --CUDASOAintegrate on +devices 7 2>&1 > gpu7
wait
./ns_per_day.py gpu0
./ns_per_day.py gpu1
./ns_per_day.py gpu2
./ns_per_day.py gpu3
./ns_per_day.py gpu4
./ns_per_day.py gpu5
./ns_per_day.py gpu6
./ns_per_day.py gpu7

3.How to compute Figure of Merit (FOM)

From the completed Workloads list, select your completed NAMD3 job. Select the STDOUT tab to view stdout logs. The FOM for NAMD3 8-GPU workload is the sum of the eight Nanoseconds per day numbers.

stmv benchmark

Copy the following code block in the Run Script text box to run stmv benchmark concurrently on all GPUs on an 8-GPU compute node.

cd /examples
BENCHMARK=/examples/stmv/stmv.namd
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 0 --CUDASOAintegrate on +devices 0 2>&1 > gpu0 &
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 8 --CUDASOAintegrate on +devices 1 2>&1 > gpu1 &
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 16 --CUDASOAintegrate on +devices 2 2>&1 > gpu2 &
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 32 --CUDASOAintegrate on +devices 3 2>&1 > gpu3 &
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 64 --CUDASOAintegrate on +devices 4 2>&1 > gpu4 &
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 72 --CUDASOAintegrate on +devices 5 2>&1 > gpu5 &
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 80 --CUDASOAintegrate on +devices 6 2>&1 > gpu6 &
/opt/namd/bin/namd3 $BENCHMARK +p1 +pemap 88 --CUDASOAintegrate on +devices 7 2>&1 > gpu7
wait
./ns_per_day.py gpu0
./ns_per_day.py gpu1
./ns_per_day.py gpu2
./ns_per_day.py gpu3
./ns_per_day.py gpu4
./ns_per_day.py gpu5
./ns_per_day.py gpu6
./ns_per_day.py gpu7

4.How to compute Figure of Merit (FOM)

From the completed Workloads list, select your completed NAMD3 job. Select the STDOUT tab to view stdout logs. The FOM for NAMD3 8-GPU workload is the sum of the eight Nanoseconds per day numbers.