A graphic processing unit implementation for the moment representation of the lattice Boltzmann method
Marco A. Ferrari, Waine Oliveira, Alan Lugarini, Admilson T. Franco, Luiz A. Hegele
Abstract
Abstract We implement the moment representation of the lattice Boltzmann method in a graphic processing unit (GPU) environment to study the computational performance of three‐dimensional fluid flows. The moment representation of the lattice Boltzmann method (MRLBM) uses up to second‐order moments of the particle distribution function to regularize and reconstruct it. By adopting this method, there is a reduction in memory usage and global memory transfer and, therefore, increased performance. The results show an increase of up to 40% in speed compared to population schemes. The validation results show there was no loss of accuracy when using single precision for turbulent flows. The adoption of single‐precision also enables higher processing speed for GPUs with low double‐precision capability and reduces the memory footprint, which can further increase performance. The implementation here presented provides a fast and straightforward environment for fluid simulation by combining single‐precision and the MRLBM.