Parallelization of LBM Code Using CUDA Capable GPU Platform for 3D Single and Two-Sided Non-Facing Lid-Driven Cavity Flow

Abstract

In this paper, a lattice Boltzmann model is developed and then parallelized employing a Compute Unified Device Architecture (CUDA) capable nVIDIA GPU platform. Numerical algorithms are developed for the solution of 3D single and two-sided non-facing lid-driven (TSNFL) cavity flow for Re = 10–1000. The algorithms are verified by solving both steady and unsteady 3D cavity and 3D TSNFL flow problems. Excellent agreement is obtained between numerical predictions and results available in literature. The results show that the CUDA-enabled LBM code is computationally efficient. It is observed that the implementation of LBM on a GPU allows at least thirty million lattice updates per second for 3-D lid driven cavity flow. Computations have been carried out for a 2-D lid driven cavity flow too. It is revealed that LBM-GPU calculation achieves 641 million lattice updates per second for the 2-D lid driven cavity flow.