On p33 of my UKCA IV book I ran mi-ah876, which fullChemN96, for 5 days
which suggests it's best to run with one thread, which was 26.3% faster.
The job for doing this is mi-ah609 and mi-aj440 (a copy of mi-ah609).
| NMPPE*NMPPN (total) | Model run length | Total time (s) | Total time on IBM (Cray/IBM scaling) | Speed (model years per day) |
|---|---|---|---|---|
| 16 * 16 (256) | 5 days | 7,417 | 5,896 (*1.26) | 0.162 |
| 16 * 32 (512) | 5 days | 3,612 | 3,012 (*1.20) | 0.332 |
| 32 * 32 (1,024) | 5 days | 1,992 | 1,843 (*1.08) | 0.602 |
| 48 * 32 (1,536, XCE) | 5 days | 1,719 | N/A | 0.698 |
| 64 * 32 (2,048) | 5 days | 1,358 | 1,421 (0.956) | 0.884 |
| 64 * 46 (2,944) | 5 days | 1,087 | N/A | 1.10 |
| 80 * 46 (3,680, XCE) | 5 days | 1,065 | N/A | 1.13 |
| 64 * 64 (4,096) | 5 days | Error message: Too many processors in the North-South direction ( 64) to support the extended halo size ( 7). Try running with 46 processors. | ||
The job for doing this is mi-ah651
| NMPPE*NMPPN (total) | Model run length | Total time (s) | Total time on IBM (Cray/IBM scaling) | Speed (model years per day) |
|---|---|---|---|---|
| 8 * 16 (128) | 10 days | 3,740 | 3,499 (*1.07) | 0.642 |
| 16 * 16 (256) | 1 month | 5,782 | 5,671 (*1.02) | 1.25 |
| 32 * 16 (512) | 1 month | 3,551 | 3,447 (*1.03) | 2.03 |
| 32 * 28 (896) | 1 month | 2,562 | 2,611 (*0.981) | 2.81 |
| 48 * 28 (1,344) | 1 month | 1,968 | 2,381 (*0.827) | 3.66 |
The job for doing this is mi-ai304
| NMPPE*NMPPN (total) | Model run length | Total time (s) | Total time on IBM (Cray/IBM scaling) | Speed (model years per day) |
|---|---|---|---|---|
| 8 * 16 (128) | 10 days | 3,197 | 3,072 (*1.04) | 0.751 |
| 16 * 16 (256) | 1 month | 4,732 | 4,921 (*0.962) | 1.52 |
| 32 * 16 (512) | 1 month | 2,866 | 3,058 (*0.937) | 2.51 |
| 32 * 28 (896) | 1 month | 2,009 | 2,282 (*0.880) | 3.58 |
| 48 * 28 (1,344) | 1 month | 1688 | 2,083 (*0.810) | 4.27 |
