Comparison of Xeon Phi and Kepler GPU Performance for Finite Element Numerical Integration

We consider two recently introduced massively multi-core architectures designed for high performance computing, the Xeon Phi coprocessor and Kepler graphics processor. We discuss the OpenCL programming model, as one that allows to look at the platforms in a unified way and to construct efficient algorithms for both of them. As an example application we investigate a typical algorithm employed in finite element codes for numerical integration. We create kernels implementing the algorithm for the two considered platforms and compare the performance obtained.