Code scaling
- Details
- Last Updated on Thursday, 08 December 2016 19:26
Version 1.4 (1D slab parallelization)
Scaling measured using IBM Power 6 (SP6 at CINECA, 2011)
| N.cores | Computing time (s) | Relative Scaling | N.cores | Computing time(s) | Relative Scaling |
| 1024³, Navier-Stokes equations | Mesh size: 1024² x 2048, Navier-Stokes solution | ||||
| 64 | 264.8 | 64 (assumed) | 128 | 280.7 | 128 (assumed) |
| 128 | 131.6 | 128.7 | 256 | 142.2 | 252.5 |
| 256 | 63.1 | 268.7 | 512 | 68.7 | 522.9 |
| 512 | 34.2 | 495.5 | 1024 | 36.3 | 989.7 |
| 1024 | 17.3 | 979.6 | Mesh size: 1024² x 2048, Boussinesq equation |
||
| 2048³, Navier-Stokes equations | 128 | 370.5 | 128 (assumed) | ||
| 512 | 716.1 | 512(assumed) | 256 | 189.5 | 250.3 |
| 1024 | 362.2 | 1012.2 | 512 | 94.9 | 499.8 |
| 2048 | 207.1 | 1771.1 | 1024 | 50.4 | 940.4 |
Version 1.7 (2D pencil parallelization)
Scaling measured on: Curie Thin nodes (at GENCI-CEA), IBM BlueGene/Q (Fermi at CINECA)
