CUDA-based parallel accelerated fluid‒thermal coupling simulation on unstructured grids

Compared with the central processing unit (CPU), the graphics processing unit (GPU) exhibits superior floating-point computation capabilities and higher memory bandwidth, making it more suitable for computationally intensive tasks in computational fluid dynamics (CFD). Fluid‒thermal coupling plays a substantial role in the aerospace industry and is an important subject in CFD. This study designs a parallel framework for the in-house CPU serial code for a coupling simulation on unstructured grids of 3D heat conduction and 1D fluid flow using compute unified device architecture (CUDA). The parallel program is developed and applied to a typical test case. The implementation achieves a maximum overall speedup ratio of 46.5 times. In addition, a data coalescing strategy is developed for unstructured grids, which reduces the time required for data transfer between the host and device to 1/300 that of the previous method. In addition, the mesh reordering method is employed and further improves the parallel program’s speedup ratio by an additional 2.1 times.