Student cluster competition 2017, team Tsinghua University: Reproducing vectorization of the tersoff multi-body potential on the Intel Skylake and NVIDIA Volta architectures.

•We reproduce the results with provided inputs of the SC’16 paper, “The Vectorization of the Tersoff Multi-Body Potential: An Exercise in Performance Portability”, on Intel Skylake, NVIDIA Volta, and CycleCloud platform with Intel Haswell.•We find the unexpected relative difference in total energy with mixed and double precision solvers when using more MPI processes, and prove it to be neglectable.•We find the dramatic performance reduction on Porter’s input from “Empirical bond-order potential description of thermodynamic properties of crystalline silicon”, and discuss the communication bottleneck that causes this issue.•We implement the suggestion in the Tersoff paper utilizing AVX512-CD to optimize the Newton force accumulation and find an overall 1% performance drop.