Nek5000/RS Performance on Advanced GPU Architectures

Abstract

60439. For information about Argonne and its pioneering science and technology programs, see www.anl.gov. DOCUMENT AVAILABILITYOnline Access: U.S. Department of Energy (DOE) reports produced after 1991 and a growing number of pre-1991 documents are available free at OSTI.GOV (http://www.osti.gov/), a service of the US Dept. of Energy's Office of Scientific and Technical Information.

Multiphysics Simulations of MSRE with NEAMS Thermal Hydraulics Tools

“…The Open Concurrent Compute Abstraction provides backends for CUDA, HIP, OpenCL, and DPC++, for performance portability across all the major GPU vendors [9]. We note that, at 80% parallel efficiency, which corresponds to 2-3 million points per MPI rank, NekRS is running at 0.1 to 0.3 seconds of wall clock time per timestep, which is nearly a 3x improvement over production runs of Nek5000 at a similar 80% strong-scale limit on CPU-based platforms [11].…”

HTTF Benchmarking Activities in FY 2023

“…The Open Concurrent Compute Abstraction provides backends for CUDA, HIP, OpenCL, and DPC++, for performance portability across all the major GPU vendors [8]. We note that, at 80% parallel efficiency, which corresponds to 2-3 million points per MPI rank, NekRS is running at 0.1 to 0.3 seconds of wall clock time per timestep, which is nearly a 3x improvement over production runs of Nek5000 at a similar 80% strong-scale limit on CPU-based platforms [9]. Nek5000 was originally developed for simulating turbulent flows with very high fidelity, i.e., DNS and LES.…”