| Test | Option| Average value, GB/s |
|-------|----------|----------------|
| GPU memory | STREAM triad | 1275 |
| Host to Device, Device to Host transfers | Pinned memory | 18..23 |
| Host to Device, Device to Host transfers | Pageable memory | 3..7 **(BAD!!!)** |
| Host to Device, Device to Host transfers | Pageable memory, huge pages | 11..13 **(BAD!)** |
| MPI point-to-point (GPU memory) | Intranode, unidirectional | 32..47 |
| MPI point-to-point (GPU memory) | Intranode, bidirectional | 23..28 |
| MPI point-to-point (GPU memory) | Across 2 nodes, unidirectional | 11..11.5 |
| MPI point-to-point (GPU memory) | Across 2 nodes, bidirectional | 10..10.8 |
| MPI point-to-point (CPU memory) | Across 2 nodes, unidirectional | 9.5..11.2 |
| MPI point-to-point (CPU memory) | Across 2 nodes, bidirectional | 8.6..9.2 **(BAD!)** |

Made with https://earth.bsc.es/gitlab/ces/hpc-for-es-team/lumi-g-test.git

8 ranks on node, transfering simultaneously, figures given per-rank