Performance Evaluation of the MODYLAS Application on Modern Multi-core and Many-Core Environments

摘要

Molecular dynamics (MD) simulations are an essential tool for various fields of science. To broaden the range of applications of MD simulations, faster and more large-scale simulations that achieve the same accuracy are required. Authors have developed an MD application named MODYLAS over several years. In our previous work, we focused on the pairwise additive calculation of potentials and forces, which is one of the hot spots in MD simulations. We have proposed new thread-level algorithms and evaluated the performance using the FX100 supercomputer system. In this study, we measure the performance of the above algorithms on Skylake-SP and Knights Landing processors and compare the results with FX100. Using the obtained results, we discuss the application performance on this hardware and the potential performance improvement by using the auto-tuning technique.