Author(s)

TIM BASHFORD, KELVIN DONNE, ARNAUD MAROTIN & ALA AL-HUSSANY

Abstract

The Dual Reciprocity BEM (DRBEM) and the Time-Dependent BEM (TDBEM) are considered in the context of radiative and time-dependent thermal transport, respectively. In order to achieve sensible solution times for realistic 3D problems with large meshes, a range of optimisation techniques are considered, and a number of parallelisation techniques applied: shared memory using multi-core threading, Graphics Processing Unit (GPU) acceleration using CUDA, and distributed memory on a high performance cluster using MPI. Particular consideration is given to practical methods to invert large dense matrices

Keywords

BEM, CUDA, MPI, threading