Foundation and analysis of computational efficiency for hydrodynamic model based on GPU parallel algorithm
Zhao Xu-don · Dalian Ligong Daxue xuebao · 2014
Unstructured grid has been widely used to establish hydrodynamic models.To rapidly get the calculation results without a cluster when the number of computational grids is too large,a highperformance computing technology which is based on GPU(graphic processing unit),is adopted to design a parallel algorithm and establish a 2D unstructured grid hydrodynamic model on CUDA(compute unified device architecture)development platform.Through the comparisons to the computational efficiency of a cluster and the Graphic Card of GTX460,the advantages of GPU method are confirmed:the speedup ratio can reach more than 10times and maintain a high growth as the increase of the number of computational grids.It is suitable for the numerical simulation of largedomain hydrodynamic models.