The results of optimising the evaluation of the Van Der Waal's forces are shown in table [
].
Table:
Timing comparison of different runs with Bak on XT4, XE6 and optimised vdw_forces on XE6
Nb. Procs |
Bak XT4 |
Bak XE6 |
Opt vdw |
16 |
199.154 |
218.722 |
212.913 |
32 |
106.790 |
98.113 |
94.590 |
64 |
63.129 |
57.494 |
56.761 |
128 |
42.036 |
39.360 |
35.121 |
256 |
27.471 |
29.492 |
25.631 |
512 |
22.137 |
25.951 |
21.641 |
|
The optimised version of vdw_forces shows a speed up of in average over all runs.A peak is observed at speed up compare to the Bak version on XE6 for 512 processes run.
Table:
Variation rate of different runs of optimised vdw_forces on XE6 with Bak on XT4 and XE6
<#1177#> |
Opt vdw comp. to |
|
|
Bak XT4 |
Bak XE6 |
16 |
6.91 |
-2.66 |
|
32 |
-11.42 |
-3.59 |
|
64 |
-10.09 |
-1.27 |
|
128 |
-16.45 |
-10.77 |
|
256 |
-6.7 |
-13.09 |
|
512 |
-2.24 |
-16.61 |
|
Average |
-6.67 |
-8.00 |
|
|
Valène Pellissier 2011-08-24