Acceleration of the space–time boundary element method using GPUs

Abstract

Matrices arising from the space–time boundary element method for the heat equation are dense and are global in space and time. Thus, they require a large amount of memory which may pose a problem when accelerating the code using GPUs. In this paper, we present a method that overcomes this issue by assembling elements of system matrices only when needed during matrix–vector multiplication. Although this requires a significantly larger amount of floating-point operations when the matrix is repeatedly applied, due to the large processing power of modern GPUs, we are still able to achieve a significant speedup compared to the original CPU code.