Parallel Matlab implementation of the lattice boltzmann method on GPUs

Abstract

The advent of cheap massively parallel computer architectures in accelerators such as GPUs provides platforms for fast high fidelity computational modelling and motivates modifications to existing softwares to take advantage of these architectures. MATLAB is one of the most widely used mathematical computing environments in technical computing. The recent advances in computer architectures has resulted in development of parallel MATLAB, which allows one to execute algorithms on multi-core and GPU enabled high performance computing platforms. The Lattice Boltzmann Method (LBM) defined on two- or three- dimensional discrete structured lattices to model transport processes has proven to be a reliable flow solver in recent years. The implementation of LBM for computing the flow-field past a NACA 63415 airfoil at low Reynolds number using Parallel MATLAB computation on three different NVIDIA GPU cards, i.e. C2070, GTX 680 and K20 is addressed in this work to set the stage for intensive large scale computation in the near future. The computational gains from the implementation of this method using Parallel MATLAB on different GPU cards are compared for varying lattice sizes and computed aerodynamic characteristics compared with the predictions from continuum based flow solvers to validate the accuracy of the computed results after parallelization.