The system simulated is PbO with 24 k-points.
This simulation involves Hartee-Fock exchange calculations. Optical properties are also examined.
Table:
Scaling of Test Case 4.
Test Case 6
Cores
Time (secs)
Speedup
VASP 5.2.2
256
14674.49
1
KPAR=2
512
7578.614
1.936
KPAR=3
768
4979.435
2.947
KPAR=4
1024
3790.061
3.871
KPAR=6
1536
2552.615
5.749
KPAR=8
2048
1944.401
7.548
KPAR=12
3072
1399.796
10.953
KPAR=24
6144
1053.134
13.93411
The original code failed to run on 512 cores or more. Hence the only way to run efficiently this problem is to employ the k-point parallelized version. This runs efficiently for this test case on up to 3072 cores.
Figure 4:
Speedup of Test Case 4 (where Speedup is taken to be 1 for 256 cores.)