ResultsWe ran each file 3 times: First using just the blade750 as a master and only using a single CPU. Second, using just the blade750 but with both CPU's. Finally, in distributed mode with three slaves machines and one CPU of the blade750 available to chop up tiles. The respective times are shown below: |
| Files | 1 CPU Elapsed Time mm:ss |
Tiles/Sec | 2 CPUs Elapsed Time mm:ss |
Tiles/Sec | Distributed Elapsed Time mm:ss |
Tiles/Sec |
| B006 |
59:03 | 0.84 | 37:56 | 1.3 | 20:14 | 2.46 |
| B008 |
77:05 | 1.20 | 47:43 | 1.9 | 32:31 | 2.82 |
| B017 |
159:56 | 0.15 | 93:10 | 0.25 | 46:07 | 0.50 |
| B012 |
71:31 | 0.70 | 42.49 | 1.16 | 26:10 | 1.90 |
| B013 |
126:30 | 0.14 | 72:33 | 0.24 | 53:14 | 0.33 |
|
Note: each GDSII file represents a chip layout. We selected one layer for tiling, typically the layer that appeared to be M1 since this is normally a fairly complex layer. The data in each tile is unionized which does put some load on the slave machines. |
| Page: 1 | 2 | 3 | 4 | [5] |