Implementation Of NS Equations In Parallel On GPU

Recently, with the great enhancement of GPU's performance, more and more industries carry out GPGPU(General-Purpose Computation on Graphics Processing Unit). In the aviation field, CFD computing accounts for plentiful time,many experts abroad take advantage of GPU to solve NS equations in parallel. Compared with cluster computing, the speedup achieved by GPU is quite large, almost 30. And the revolution of CFD computing resulted from GPU is encouraging. Many CFD commercial software such as ANSYS, Autodesk involve in the research of GPU. Meanwhile, in our country, there are few cases which could illustrate that solving NS equations in parallel on GPU is feasible and can gets a high speedup, not to mention that in the engineering practice. This situation blocks the development of solving NS equations on GPU.According to the situation, this paper devises and accomplishes the algorithm of solving NS equations in parallel on GPU, which aims to introduce the advanced computing techniques and promote the power and techniques of CFD computing. This paper focus on the following works: (1) Investigate the existing numerical method of solving NS equations for the viscous compressible flows; (2) devise and accomplish the algorithms of solving NS equations in parallel on GPU, and optimizes the application performance with many techniques; (3) compare the serial solution and parallel solution of the ORENA M6 wing and ANOTHER wing,Compared the parallel solution to the serial solution of ORENA M6 wing and ANOTHER wing, we can conclude that the algorithm of solving NS equations in parallel on GPU is correct. On GPU, a speedup of 24 is got for the ORENA M6 wing on a grid of 244800(51×30×160)meshes, and a speedup of 37 for ANOTHER wing on a grid of 1676480(208×65×124) meshes. The result demonstrates that solving NS equations in parallel on GPU could achieve a high speedup with nice scalability.