Parallel Solver For Incompressible Viscous Flow Using CUDA

Due to the non-linearity of the equations, Computational fluid dynamics is very time-consuming,and often need to deal with large-scale data.The researchers faced two challenges,one is it cost too much time,the other is the large amount of memory usage.Fine simulation of some flow problems still can not be achieved even on a massively parallel machine.Therefor,computational fluid dynamics has posed a severe challenge to high-performance parallel computing.Recent years,the performance of GPU improved quickly.compared with CPU,it has distinct advantages in the processing capacity and memory bandwidth.GPU multi-thread design is very suitable for parallel computing.but do not need to pay too much price in terms of price and power consumption.GPU provides a new solution for high-performance parallel computing.After the release of the first version of CUDA.researchers can write programs with their proficiency programming language such as C, Fortran, without knowing graphics APIs.Currently, GPU computing has become a cutting-edge research direction of computational fluid dynamics.The content of this dissertation mainly includes for parts.as follows:In the first part.the key points of CUDA programming are introduced in this paper,through the systematic study of CUDA. CUDA C parallel version program for Jacobi iteration was designed,and it got a good compute acceleration.In the second part.CFD boundary condition treatment are discussed with CUDA.Three processing modes was designed.the computation speed of each model was compared bu experiment.The result show that When the flow field data scale is much larger than the size of the edge points, Independently design kernel functions which compute the flow area and boundary can maintain the program structure clear and can achieve the optimal computation speed.In the third part,a solving two-dimensional incompressible flow solver was designed based on artificial compression method and finite volume scheme.and it was used to simulate2D cavity flow.The result has high accuracy.and got more than10 times acceleration than the serial C program.In the last part,the CUDA based Lattice Boltzmann Method of CFD was used to implement the simulation of2D cavity flow.and obtained15times acceleration than the serial program.