A Performance Study of a Dual Xeon-Phi Cluster for the Forward Modelling of Gravitational Fields

Table 1

Computation times when using the Xeon-Phi coprocessors. The best results are obtained when 236 threads are created per coprocessor; it means that we are obtaining a gain using the multithreading (4 threads per core), but when the number of threads is increasing, only an overhead is created.

Threads	Single Xeon-Phi	Dual Xeon-Phi
59	1 h 32 m 31 s	0 h 47 m 24 s
118	0 h 55 m 52 s	0 h 28 m 49 s
177	0 h 44 m 09 s	0 h 22 m 49 s
236	0 h 37 m 24 s	0 h 18 m 46 s
295	0 h 48 m 30 s	0 h 24 m 45 s
354	0 h 44 m 26 s	0 h 22 m 43 s
413	0 h 41 m 27 s	0 h 21 m 03 s
472	0 h 38 m 02 s	0 h 19 m 19 s