Development of GPU-based Parallel CFD Solver for Unstructured Meshes

As computer technology and numerical algorithms rapidly develop, the role of computational fluid dynamics (CFD) in the fields of mechanical and aerospace engineering has become increasingly prominent. However, the outdated hardware level has hindered the application of CFD in both theory and engineering. In this article, we developed a CFD solver based on unstructured grids for compressible hypersonic flow. This solver achieved parallel computing based on GPUs using the CUDA programming model. We optimized the code to cater to the characteristics of CPU/GPU heterogeneous systems. The solver used a second-order finite volume method and an adaptive hybrid Riemann solver to accurately and robustly solve hypersonic flow problems. For grid sizes Up to tens of millions, the solver can achieve two orders of magnitude speedup.