Student Cluster Competition 2017, Team University of Utah: Reproducing Vectorization of the Tersoff Multi-Body Potential on the Intel Broadwell and Intel Skylake Platforms

Abstract

Abstract This paper describes the University of Utah SC17 Student Cluster Competition team’s efforts to reproduce the results of an SC16 paper titled “The Vectorization of the Tersoff Multi-Body Potential: An Exercise in Performance Portability”. This application optimizes the Tersoff potential within the widely-used molecular dynamics code LAMMPS. Tersoff multi-body potentials compute the force between two particles as a function of their distance and the relative position of surrounding particles. The application described in the cited paper focuses on providing node level and SIMD parallelization strategies for the Tersoff multi-body potential that are portable to different CPU instruction sets and to GPUs. In this paper we collect measurements on our SC17 cluster, which consists of a Broadwell head node and Skylake compute nodes. Our testing demonstrates that we can produce similar speedups to the original publication on the AVX2 instruction set on both the Broadwell and Skylake platforms. Additionally, we examined the performance on the Skylake node using the AVX-512 instruction set.