Nek5000/RS Performance on Advanced GPU Architectures

TL;DR

This paper evaluates the performance of NekRS, a GPU-accelerated CFD code, on various advanced GPU architectures, demonstrating its scalability and efficiency for exascale simulations in nuclear reactor applications.

Contribution

It provides the first comprehensive performance analysis of NekRS across multiple cutting-edge GPU platforms, highlighting its scalability for exascale computing.

Findings

NekRS achieves high strong-scaling efficiency on multiple GPU architectures.

Performance results demonstrate suitability for large-scale nuclear reactor simulations.

NekRS outperforms previous CPU-based implementations in key metrics.

Abstract

We demonstrate NekRS performance results on various advanced GPU architectures. NekRS is a GPU-accelerated version of Nek5000 that targets high performance on exascale platforms. It is being developed in DOE's Center of Efficient Exascale Discretizations, which is one of the co-design centers under the Exascale Computing Project. In this paper, we consider Frontier, Crusher, Spock, Polaris, Perlmutter, ThetaGPU, and Summit. Simulations are performed with 17x17 rod-bundle geometries from small modular reactor applications. We discuss strong-scaling performance and analysis.