Comparison of acceleration methods of matrix calculations in embedded systems

Abstract

In today's algorithms for sound localization techniques, matrix calculations are ubiquitous. Therefore, this work deals with the analysis of matrix calculations and their possible realization on embedded systems. For this purpose, common acceleration technologies such as processors, GPU processing and acceleration with the help of FPGAs are compared. The results show that a graphics chip is capable to accelerate such a matrix vector multiplication compared to an implementation on a processor. Therefore a runtime of an implementation on an FPGA cannot be achieved by a GPU.