Performance of a Code Migration for the Simulation of Supersonic Ejector Flow to SMP, MIC, and GPU Using OpenMP, OpenMP+LEO, and OpenACC Directives

Figure 13

Computing times and the corresponding speed-up factors compared to Amdahl’s law with and . As can be seen, the Xeon Phi stops scaling in two threads per core for all the proposed designs (for this domain size). Unlike the conventional CPU the Phi does not show improvements with the vectorization, due to the decrease of vectorization effects with many cores; that is, there are no sufficient floating point operations per core.

(a) Computing time

(b) Speed-up