E-mail from Julien on 2 October 2015 gives
| (NEMO East-West)*(NEMO North South) (total) | Model run length | Total time (s) | Total time on IBM (Cray/IBM scaling) | Speed (model years per day) |
|---|---|---|---|---|
| 16x12 (192) | 10 days | 5,407 | 6,322 (*0.855) | 0.444 |
| 32x12 (384) | 10 days | 3,042 | 2,760 (*1.10) | 0.789 |
| 32x20 (640) | 10 days | 1,950 | 1,578 (*1.24) | 1.23 |
| 32x30 (960) | 10 days | Failed | 1,082 | |
| 36x30 (1,080) | 10 days | 1,213 | Estimate 1,002 (2.394 days/year) (*1.21) | 1.98 |
Richard has run his vn3.4 code for 10 days, see his notes on this.
| Number of cores | Model run length | Duration (s) | Expected duration for one model year (days) | Expected duration on IBM (Cray/IBM scaling) | Speed (model years per day) |
|---|---|---|---|---|---|
| 96 | 10 days | 283 | 0.118 | 0.122 (*0.967) | 8.47 |
| 192 | 10 days | 149 | 0.0621 | 0.0670 (*0.927) | 16.1 |
| 288 | 10 days | 108 | 0.0450 | 0.0486 (*0.926) | 22.2 |
| 320 | 10 days | 99 | 0.0413 | 0.0449 (*0.920) | 24.2 |
Suprisingly, this run seems to be quicker on Cray than IBM.
