GPU Implementation of a Lattice Boltzmann Flow Solver

In this paper the first national implementation of a Lattice Boltzmann flow solver on Graphical Processing Unit (GPU) for fast computation of fluid flow is reported. GPUs are multi-processor devices which can handle thousands of operations in parallel. They have very recently become available for general purpose scientific computations. For fluid flow simulations the Lattice Boltzmann Method (LBM), an alternative to the conventionalNavier-Stocks solvers, is used. LBM is a well known method for fluid flow simulations which, due to the explicit and local nature of its computations, can be effectively parallelized. We have developed a CUDA code for the LBM simulation of incompressible fluid flow inside a two dimensional square cavity. Fluid flow simulations are performed at Reynolds Numbers of 1000 and 5000. A comparison between the CPU and theGPU codes shows that the GPU code could run more than 180 times faster . It is also shown that the single precision LBM results are as accurate as those of a double precision model, which further justifies the use of GPUs for lattice Boltzmann simulation of fluid flows

کلیدواژه ها:

CFD ، Parallel Computing ، GPU ، LatticeBoltzmann Method