Algorithms performance evaluation in Hybrid systems

Abstract

Over recent years a new branch of parallel computing has emerged: hybrid architectures composed of CPUs (Central Processing Unit) and GPU (Graphics Processing Unit). The objective of this work is to quantify the dimension of the performance gains comparing sequential processing of known algorithms with parallelized processing offered by hybrid architecture. It was used OpenCL (Open Computing Language) generated by the AMD Aparapi Framework to implement popular computational algorithms. Results show the performance using Aparapi Framework and the conclusions details the expected scale of algorithm speed up comparing with other works found in the literature using pure OpenCL library.