| ... | ... | @@ -102,7 +102,7 @@ Finally, for all these strategies, we can also think about moving from the class |
|
|
|
|
|
|
|
Comparison of the update phase, without considering task overlapping. We will measure the time elapsed from the end of the backward pass to the start of the forward pass. No optimization of MPI_SUM has been done yet
|
|
|
|
|
|
|
|
| Version | Number of nodes | Update phase time |
|
|
|
|
| Version | Number of ranks | Update phase time |
|
|
|
|
| ---------- | --------------- | ----------------- |
|
|
|
|
| Strategy 1 | 5 |510 ms |
|
|
|
|
| Strategy 2 | 4 |210 - 230 ms |
|
| ... | ... | |
