CUDA Implementation of a Lattice Boltzmann Method and Code Optimization
Tata Consultancy Services
@article{jaincuda,
title={CUDA Implementation of a Lattice Boltzmann Method and Code Optimization},
author={Jain, Anubhav}
}
We study fluid flow in a 2D lid driven cavity for large Reynolds numbers using multirelaxation time – Lattice Boltzmann Method(LBM). LBM is an alternative to conventional CFD methods that solve Navier-Stokes equations to simulate incompressible fluid dynamics. In LBM, one solves the linearized Boltzmann equation on a discrete lattice to study spatio-temporal evolution of flow field. The data parallel implementation of the Lattice Boltzmann Method makes the GPGPU as a platform of choice for such computation. Several CUDA optimizations are implemented to achieve desired performance, these are discussed below.
March 22, 2014 by hgpu