Development and optimization of the nekRS CFD solver on accelerated platforms.

PI Name Kristopher Rowe
PI Institution Argonne National Laboratory
Collaborating ANL Division Leadership Computing Facility (LCF)
Project Description

NekRS is a reimplementation of the Nek5000 CFD solver for exascale platforms and is a mission critical application for both ALCF and other divisions at Argonne. While ongoing development of NekRS continues at a rapid pace, testing and optimization across the wide range of hardware which the application must be supported on remains a challenge. The broad collection of hardware contained in the JLSE testbeds make JLSE a unique and ideal platform to support NekRS development.

The purpose of this project is the following:
1. To establish a comprehensive set of CI pipelines for testing the NekRS code on NVIDIA, AMD, and Intel GPUs.
2. To perform a cross-platform study of NekRS on NVIDIA, AMD, and Intel GPUs--examining the performance, comparing the strengths and weaknesses of each platform, and identifying areas for improvement.
3. To port NekRS to new testbeds as they become available, conducting similar performance analysis.

Ultimately, the knowledge gained through this work will help to better support ALCF users and improve efficient use of ALCF resources by INCITE and ALCC projects which depend on the NekRS code.

Testbed

gpu_a100, gpu_h100, gpu_gh200, gpu_amd_mi250, florentia (pvc)