Simulation of thermal flows by lattice Boltzmann method on the CUDA computational platform

Abstract

Originating from lattice gas automata theory, the lattice Boltzmann method (LBM) is an interesting alternative to the solving of Navier-Stokes equations. In contrast to isothermal simulations, for a while thermal flow simulations were challenging for LBM. Thermal flow simulations are important task in various fields of research. Despite a large amount of work and research the dynamics of thermally induced flows are still highly demanded. Motivation of this work is development of computational tool for simulation of the dynamics of thermal flows. To this purpose, we developed LES-LBM solver accelerated by the Graphics Processing Unit (GPU) on the CUDA computational platform, integrating LBM with Large Eddy Simulation (LES). Simplicity of coding is usually an appealing feature of the LBM. Conventional implementations of LBM suffer from high memory consumption and poor computational performance. The main advantage of the solvers based on GPU is their ability to perform significantly more floating point operations per unit time (FLOPS) than a Central Processing Unit (CPU) and a good scalability of explicit parallel algorithms. LES-LBM code was tested on the NVIDIA GeForce GTX 1050 ti and NVIDIA TESLA K80 GPUs.