In this part, all the cases are taken in account.
One can observe that the speedup is slightly greater on average than the no frozen combination routines runs, a surprising result but the differences are small and when the above runs were performed large variations in runtime were observed. On average, the best combination is faster than the Bak version on XT4, and faster than the one on XE6. A peak is reached for 512 processes, with speedup compare to XE6. Another one is noticed for 128 cores with a time reduction of compare to XT4.
|
Valène Pellissier 2011-08-24