Bonus topic: Parallelization#
Here is a benchmark of using multiple nodes along with k-point parallelize division on a 4x4x1 HB sheet. The supercomputer used is Ohtaka ISSP so the cpu configuration is 64 cores per cpu and 2 cpu per node for a total of 128 cores per node.
Node |
NK |
CPU Time |
Wall Time |
|---|---|---|---|
1 |
1 |
1m46.90s |
2m 1.68s |
1m47.75s |
1m57.91s |
||
1m52.00s |
1m58.97s |
||
1 |
2 |
1m33.13s |
1m42.31s |
1m32.42s |
1m39.14s |
||
1m35.88s |
1m42.57s |
||
2 |
1 |
1m25.44s |
1m43.07s |
1m26.65s |
1m32.39s |
||
1m22.56s |
1m33.72s |
||
2 |
2 |
1m 7.67s |
1m12.38s |
1m 4.57s |
1m 8.69s |
||
1m 0.98s |
1m 5.67s |
||
5 |
5 |
40.42s |
51.43s |
30.56s |
34.65s |
||
34.67s |
37.14s |