Development of Triangular Equilibrium Flux Method Using Multiple Graphics Processing Unit Acceleration with MPI for the Euler Equations

  • 蘇 柏勝

Student thesis: Master's Thesis

Abstract

Kinetic-theory based Finite Volume Methods (FVM) have become increasingly common in Computational Fluid Dynamics (CFD) Many of these methods focus on computation of fluxes through integrating a velocity probability distribution function such as the early Equilibrium Flux Method (EFM) However analytical computation of these moments results in the exponential function and error function which is not only time-consuming but may also introduce error through truncation The purpose of this study is to investigate an alternative – the Triangular Equilibrium Flux Method (TEFM) which uses the sum of simpler distribution functions to approximate the Maxwell-Boltzmann equilibrium velocity probability distribution function – applied to parallel computing using MPI with a specific focus on the A-series of AMD APU’s To further enhance the performance the hybrid MPI-CUDA parallelization paradigm is employed with multiple GPUs to achieve higher simulation efficiency Furthermore several benchmarks are used to verify both first and second order numerical results
Date of Award2016 Jul 21
Original languageEnglish
SupervisorMatt-Hew Smith (Supervisor)

Cite this

'