A new approach to the lattice Boltzmann method for graphics processing units

Emerging many-core processors, like CUDA capable nVidia GPUs, are promising platforms for regular parallel algorithms such as the Lattice Boltzmann Method (LBM). Since global memory on graphic devices shows high latency and LBM is data intensive, memory access pattern is an important issue to achieve good performances. Whenever possible, global memory loads and stores should be coalescent and aligned, but the propagation phase in LBM can lead to frequent misaligned memory accesses. Most previous CUDA implementations of 3D LBM faced this problem using low latency on chip shared memory. Instead of this, our CUDA implementation of LBM follows carefully chosen data transfer schemes in global memory. On the 3D lid-driven cavity test case, we obtained 2× to 3× speed-up over previously published performances, achieving up to 86% of the global memory maximal throughput. As a consequence we show that highly efficient implementations of LBM on GPU are possible, even for complex models.