Evaluation of AMD GPUs and their Applicability for CFD and ML Workloads
Hess, Andrew (US Naval Research Laboratory)
Co-Authors:
Talya Stehley
Keith Obenschain
Category:
Incorporation of GPUs/Accelerators into Physics-based Codes
Competition in the High-Performance Computing GPGPU market has emerged with GPGPUs from Advanced Micro Devices (AMD) and Intel targeting Exascale class systems. The recently released AMD GPUs such as the MI100 and MI210 have reached the stage where they are competitive for DoD relevant problems. This presentation is an assessment of both the hardware and software (ROCm) platforms for AMD GPUs, with a focus on how its capabilities are used with NRL's CFD code, JENRE. JENRE is a Discontinuous Galerkin (DG) based CFD code designed to flexibly handle complex multiphysics problems, especially supersonic and reactive flows. We will describe the porting process and how it contrasts with fielded GPU platforms. In addition, we will compare training throughput and scaling efficiencies using a standard image classification task against competing GPU platforms.