Why is matrix-vector multiplication with small matrix sizes faster on CPU (core 2 quad) than on Xeon phi?

Question

Intel

Why is matrix-vector multiplication with small matrix sizes faster on CPU (core 2 quad) than on Xeon phi?

Answer 1

If this is a separate card - because you need to send data to it and pick it up from it. This is the bottleneck of dealing with individual cards. There is another option that you need to compile a computing core for this card, which also takes time.