Performance Analysis of Parallel Processing on GPU for Simple Mathematical Computations

DOI

10.2991/aisr.k.200424.044How to use a DOI?

Keywords

parallel processing, NVIDIA GPU, CUDA, processing time, general purpose computing

Abstract

Over time, more and more data being produced. We need high-processing computing to process this big data. One solution to this problem is to implement parallel processing using a Graphical Computing Unit (GPU). Theoretically, processing mathematical computation on GPU will always be faster than CPU, because GPU has more than hundreds of Arithmetic Logical Units (ALUs) while CPU only has less than 10 ALUs. But to be able to process data on GPU, we need to explicitly transfer data from RAM to global memory of GPU. This process creates fairly high cost. In this research, we analyze performance of GPU compared to CPU for 2 mathematical computations, namely 1-dimensional vector addition and 2-dimensional matrix multiplication. From experimental result, we can conclude that for 1-dimensional vector addition, however big the data size, it is better to use CPU than GPU. In this case, cost of data transmission is more significant than acceleration of parallel computational process. For 2-dimensional matrix multiplication, if we use matrix larger than 96 x 96 floating point, it is better to use GPU than CPU.

Copyright

© 2020, the Authors. Published by Atlantis Press.

Open Access

This is an open access article distributed under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/).

Download article (PDF)

TY  - CONF
AU  - Mastura Diana MARIESKA
AU  - M. Ridho Putra SUFA
AU  - Adi WIDIANTO
AU  - Novi YUSLIANI
AU  - Rahmat Izwan HEROZA
PY  - 2020
DA  - 2020/05/06
TI  - Performance Analysis of Parallel Processing on GPU for Simple Mathematical Computations
BT  - Proceedings of the Sriwijaya International Conference on Information Technology and Its Applications (SICONIAN 2019)
PB  - Atlantis Press
SP  - 294
EP  - 299
SN  - 1951-6851
UR  - https://doi.org/10.2991/aisr.k.200424.044
DO  - 10.2991/aisr.k.200424.044
ID  - MARIESKA2020
ER  -