Direct numerical simulation of compressible turbulence accelerated by graphics processing unit. Part 1: An open-source high accuracy accelerated computational fluid dynamic software

TL;DR

This paper presents OpenCFD-SCU, an open-source GPU-accelerated CFD software for DNS of compressible turbulence, achieving over 200x speedup over CPU-based codes and supporting large-scale, high-precision simulations.

Contribution

The paper introduces a GPU-accelerated CFD software with high accuracy and scalability, significantly improving simulation speed and capacity for compressible turbulence.

Findings

GPU acceleration exceeds 200x compared to CPU software.

Supports large-scale simulations with up to 31.2 billion mesh points.

Achieves 98.7% parallel weak scalability with 24,576 GPUs.

Abstract

This paper introduces open-source computational fluid dynamics software named open computational fluid dynamic code for scientific computation with graphics processing unit (GPU) system (OpenCFD-SCU), developed by the authors for direct numerical simulation (DNS) of compressible wall-bounded turbulence. This software is based on the finite difference method and is accelerated by the use of a GPU, which provides an acceleration by a factor of more than 200 compared with central processing unit (CPU) software based on the same algorithm and number of message passing interface (MPI) processes, and the running speed of OpenCFD-SCU with just 512 GPUs exceed that of CPU software with 130\,000 CPUs. GPU-Stream technology is used to implement overlap of computing and communication, achieving 98.7\% parallel weak scalability with 24\,576 GPUs. The software includes a variety of high-precision finite difference schemes, and supports a hybrid finite difference scheme, enabling it to provide both robustness and high precision when simulating complex supersonic and hypersonic flows. When used with the wide range of supercomputers currently available, the software should able to improve the performance of large-scale simulations by up to two orders on the computational scale. Then, OpenCFD-SCU is applied to a validation and verification case of a Mach 2.9 compression ramp with mesh numbers up to 31.2 billion. More challenging cases using hybrid finite schemes are shown in Part 2(Dang, Li et al. 2022). The code is available and supported at \url{http://developer.hpccube.com/codes/danggl/opencfd-scu.git}.