Performance Evaluation and Scalability Analysis of NPB-MZ on Intel Xeon Phi Coprocessor

Abstract

Intel Many Integrated Cores (Intel MIC) is a novel architecture for high performance computing (HPC). It features large thread parallelism and wide vector processing units, targeting highly parallel applications. The HPC communities are faced with the problem of porting their applications to the MIC platforms. But it is still an open question that how current HPC applications can exploit the capabilities of MIC. This paper evaluates the performance of NPB-MZ programs which are derived from real world Computational Fluid Dynamics (CFD) applications on Intel Xeon Phi coprocessor, the first MIC product. The strong scaling behaviors of the applications with different process-thread combinations are investigated. The performance obtained on the Intel Xeon Phi coprocessors is compared against that obtained on Sandy Bridge CPU based computer nodes. The results show that these programs can achieve good parallel scalability when running with appropriate combinations of processes and threads. But their absolute performance on Intel Xeon Phi coprocessor is significantly lower than that on CPU node, due primarily to the much lower single thread performance. The findings of this paper are of help to the performance optimization of other applications on MIC.