JHU MT Wiki
The CLSP cluster has the following machines with GPUs:
| Machines | GPU/Machine | GPU Total | Type | Memory |
| b01-05,07-10 | 4 | 36 | Tesla K10.G2 | 3526 MiB |
| b06 | 2 | 2 | Tesla K20m | 4742 MiB |
| b11-18,20 | 4 | 36 | Tesla K80 | 11439 MiB |
| b19 | 2 | 2 | Tesla M40 | 22939 MiB |
| c01-11 | 4 | 44 | GTX 1080ti | 11172 MiB |
Wall time for amun / marian benchmarks
| GPU | amun | marian |
| Tesla K80 | 388 sec. | 1723 sec. |
| Tesla M40 | 195 sec. | 831 sec. |
| GTX 1080ti | 118 sec. | 522 sec. |
Amun benchmark is translating a 3000 sentence test set (nmt17-de-en).
Marian benchmark is 1000 iterations (about 80,000 sentence pairs) of training (nmt17-de-en). Nematus on Tesla K80 takes about 3,000 seconds for 1000 iterations of training (with a simpler model, i.e., not using layer normalization).