Latency and Bandwidth Microbenchmarks of Six US Department of Energy Systems in the Top500

Carl Pearson, Christopher M. Siefert, Stephen L. Olivier, Andrey Prokopenko, Timothy J. Fuller, Jonathan J. Hu
IEEE Cluster 2023
Developers of portable high-performance computing applications are often concerned with basic performance properties across a variety of systems. Although supercomputing systems are comprehensively benchmarked during their acceptance testing process, results are not publicly disseminated and comparisons are typically restricted to immediate predecessor systems. This work presents selected single-node microbenchmarks of archetypal United States Department of Energy computers present in the June 2023 Top 500 list. These systems feature Intel or AMD CPUs, including Xeon Phi, and AMD or Nvidia GPUs, and provide a reasonable reference for what users can expect from these systems.