基于投影法的GPU并行计算研究

Research on GPU Parallel Computation Based on Projection Method

在计算流体力学(CFD)领域中,随着对流动传热问题研究的深入,涉及到的问题复杂程度也不断提高,导致计算耗时大幅增加。为了快速求解复杂的流动传热问题,国内外学者结合高效的图形处理器(GPU)并行计算技术,完成了部分基于GPU并行计算的求解算法的研究。与CPU串行计算相比,GPU并行计算能够实现数倍、数十倍甚至数百倍的加速效果。为了实现非稳态流动传热问题的快速求解,开发了基于投影法的GPU并行计算策略,并以二维顶盖驱动流为例研究了GPU并行计算求解性能。研究结果表明:随着网格规模的不断增大,GPU并行计算性能发挥的更为明显,当网格数达到500×500时,单块GPU并行计算速度是CPU串行计算的28.5倍,实现了非稳态流动传热问题的快速计算。

In the field of computational fluid dynamics(CFD),with the deepening of the research on the flow and heat transfer problems,the computation time increases greatly due to the increase of complexity of the problems involved.In order to solve complex flow and heat transfer problems rapidly,the domestic and foreign scholars developed some efficient solving algorithms on the basis of GPU parallel computation,combining the efficient GPU parallel computation technology.Compared with CPU serial computation,GPU parallel computation can yield speedups of several times,dozens of times or even hundreds of times.To realize the fast solution of unsteady flow and heat transfer problems,a GPU parallel computing strategy based on projection method is proposed,and its solving performance is studied by taking two-dimensional lid-driven cavity flow as an example.It can be concluded that the advantages of GPU parallel computation are more obvious with the increase of the grid sizes.When the grid number reaches 500×500,the GPU parallel computing speed is 28.5 times higher than that of the CPU serial computation,realizing the fast calculation of unsteady flow and heat transfer problems.