中文摘要：

采用多GPU并行的格子Boltzmann方法（1attice Boltzmann method，LBM）对充分发展的槽道湍流进行了直接数值模拟．GPU（graphic processing unit）的数据并行单指令多线程（single—instruction multiple—thread，SIMT）特征与LBM完美的并行性相匹配，使得LBM求解器在GPU上运行获得了极高的性能，亦使得大规模DNS（direct numefical simulation）在桌面级计算机上进行成为可能．采用8个GPU，网格数目达到6．7×10’，全场网格尺寸△’=1．41．模拟3x10。个时间步长，用时仅24h．另外，直接模拟结果无论是在平均流速或湍流统计量上均与Moser等的结果吻合得很好，这也证实了二阶精度的格子Boltzmann法直接模拟湍流的能力与有效性．

英文摘要：

The wall-bounded turbulent flow was simulated directly （DNS） by lattice Boltzmann method （LBM） through multi-GPU parallel computing. The Data-parallel SIMT （single-instruc- tion multiple-thread） characteristic of GPU matched the parallelism of LBM well, which led to high efficiency of GPU on the LBM solver. At the same time, it brought possibility for large- scale DNS on the desk-top supercomputer. In this DNS work, 8 GPUs were adopted. The num- ber of meshes of 6.7x107, which resulted in a non-dimensional mesh size of A~ = 1.41 for the whole solution domain. It took only 24 hours for the GPU-LBM solver to simulate 3 x l06 LBM steps. As a result, both the mean velocity and turbulent variables, such as Reynolds stress and velocity fluctuations, agree well with the results of Moser, et al. The capacity and validity of LBM in simulating turbulent flow are verified.