Solving two-dimensional euler equations on GPU

The potential of GPU in scientific computation has been tested in many applications during recent years. CUDA, short for Compute Unified Device Architecture, is a solution of GPGPU proposed by Nvidia Corporation. Researchers can improve performance of programs and reduce time cost with CUDA involved. This paper presents a flow solver for two-dimensional airfoil governed by Euler equations on GPU using CUDA. Body-fitted structured mesh is employed here, and finite-volume method is used to obtain the solution. Since the number of concurrent threads supported by CUDA is very huge(more than 10¹⁵), the proposed solver uses a CUDA thread to calculate the flow variables for each cell of the mesh. Two dimensional CUDA blocks are used to manage threads in the application. This paper gives a performance comparison between GPU parallel implementation and CPU serial implementation, and the results show that the proposed method has good efficiency. As CUDA supports three dimensional thread blocks, the solver can be extended to three dimensional cases naturally.