Hybrid CPU + Xeon Phi implementation of the Particle-in-Cell method for plasma simulation

Abstract

This paper presents experimental results of Particle-in-Cell plasma simulation on a hybrid system with CPUs and Intel Xeon Phi coprocessors. We consider simulation of two relevant laserdriven particle acceleration regimes using the Particle-in-Cell code PICADOR. On a node of a cluster with 2 CPUs and 2 Xeon Phi coprocessors the hybrid CPU + Xeon Phi configuration allows to fully utilize the computational resources of the node. It outperforms both CPU-only and Xeon Phi-only configurations with the speedups between 1.36 x and 1.68 x.