Implementation of the lattice Boltzmann method on GPU clusters

An implementation of the lattice Boltzmann method for the model D3Q19 on GPU clusters in the framework of the NVIDIA CUDA technology is considered. The scalability of this implementation for the Lomonosov MSU supercomputer complex is estimated. The numerical results obtained for the steady Poiseuille flow in a tube and for evaluating the drag coefficient of a sphere are in good agreement with the corresponding analytical dependencies.