APS Logo

High-Accuracy, High-Performance DFT Calculations on Exascale CPU-GPU Architectures with the RMG code

ORAL

Abstract

The promise of Exascale computer architectures is close to being realized, with the first system operational in 2022. However, applying their computational power to the study of large-scale scientific problems requires careful attention to the interplay between system architecture and computational algorithms. RMG is a software package for electronic structure calculations designed from inception for massively parallel platforms, which has been further optimized for pre-exascale machines such as Summit at ORNL. We describe the development of algorithms to further improve accuracy, convergence, and performance, as well as the implementation of advanced features such as hybrid functionals, semi-local pseudopotentials, and spin-orbit coupling. Tests on the Crusher platform at ORNL show excellent performance and scaling, which should translate well to the exascale Frontier that has identical nodes. RMG source code and build scripts for the pre-exascale Summit, Cray XE-XK, clusters, Linux, and Windows workstations are at www.rmgdft.org, together with help files and examples. A web interface for generating RMG input from standard atomic structure files or other commonly used electronic structure codes is also available, as is an interface for analysis of RMG output.

Presenters

  • Emil Briggs

    North Carolina State University

Authors

  • Emil Briggs

    North Carolina State University

  • Wenchang Lu

    North Carolina State University

  • Jerzy Bernholc

    NC State University