Home // ICDS 2012, The Sixth International Conference on Digital Society // View article

Fast Polynomial Approximation Acceleration on the GPU

Authors:
Lumír Janošek
Martin Němec

Keywords: GPU; CUDA; Direct Memory Access; Parallel Reduction; Approximation

Abstract:
This article presents the possibility of parallelization of calculating polynomial approximations with large data inputs on GPU using NVIDIA CUDA architecture. Parallel implementation on the GPU is compared to the single thread CPU implementation. Despite the enormous computing power of today's graphics cards there is still a problem with the speed of data transfer to GPU. The article is mainly focused on the implementation of some ways of transferring data from memory into GPU memory. The aim is to show what method is suitable for a large amount of data being processed and what for the lesser amount of data. Afterwards performance characteristics of the implementation of the CPU and GPU are matched.

Pages: 69 to 72

Copyright: Copyright (c) IARIA, 2012

Publication date: January 30, 2012

Published in: conference

ISSN: 2308-3956

ISBN: 978-1-61208-176-2

Location: Valencia, Spain

Dates: from January 30, 2012 to February 4, 2012