Enabling Ginkgo as Numerics Backend in nekRS Employing A Loosely-Coupled Configuration File Concept

In computational fluid dynamics (CFD), the choice of numerical methods can significantly impact the overall simulation runtime. While it is virtually impossible to know the optimal solver plus preconditioner configuration for every hardware and application setup, it is valuable for CFD engineers to have access to and evaluate different numerical methods to customize the setup for efficient execution. In this paper, we demonstrate how the Ginkgo high-performance numerical linear algebra library is integrated as a math toolbox into the nekRS state-of-the-art computational fluid dynamics simulation library to give CFD engineers access to a plethora of solvers and preconditioners CFD engineers. Using three application test cases, we demonstrate how picking numerical methods from the Ginkgo library can accelerate simulations on supercomputers featuring NVIDIA’s Ampere GPUs and Grace Hopper superchips.