Development background of the combination of lattice Boltzmann method (LBM) and graphic processing unit (GPU) computing on a desktop computer was discussed. A detailed analysis was conducted of the standard LBM form and the reason of its natural parallelism. The CUDA programming model and the Kepler computing architecture used in the computing were expounded. In order to verify the application capacity of GPU computing of LBM on a desktop computer, numerical simulations of two-dimensional flow around a square cylinder were conducted, and simulation results with the finite volume method calculation results were compared. The results show that as for the problem of twodimensional flow around a square cylinder, computational efficiency of GPU computing is approximately 3.4 times of central processing unit (CPU) computing, which proved the computing capacity of GPU computing of LBM on desktop computers to universal scientific computing.