====== Performance of the H2O-64 Benchmark on Piz Daint ======

^ Nodes ^ Cores ^ Best time (s) ^ Config ^ Time with GPU, 8 OMP threads per MPI task (s) ^
| 1 | 8 | 210.039 | 1 OMP thread per MPI task, no GPU | 267.166 |
| 2 | 16 | 110.739 | 1 OMP thread per MPI task, no GPU | 154.773 |
| 4 | 32 | 61.16 | 1 OMP thread per MPI task, no GPU | 95.452 |
| 8 | 64 | 33.984 | 1 OMP thread per MPI task, no GPU | 76.918 |
| 12 | 96 | 31.252 | 1 OMP thread per MPI task, no GPU | 71.091 |
| 16 | 128 | 25.993 | 1 OMP thread per MPI task, no GPU | 54.976 |
| 24 | 192 | 19.885 | 1 OMP thread per MPI task, no GPU | 66.047 |
| 32 | 256 | 21.478 | 1 OMP thread per MPI task, no GPU | 56.765 |
| 64 | 512 | 20.466 | 2 OMP threads per MPI task, no GPU | 45.094 |


{{:performance:h2o-64-comparison-piz-daint-piz-daint-gpu.png?direct|}}