COMPUTER AND INFORMATION SCIENCE

ruralrompSoftware and s/w Development

Dec 2, 2013 (3 years and 6 months ago)

66 views

COMPUTER AND INFORMATION SCIENCE


A Performance Comparison of General Purpose GPU Computing, Sungbo Jung,
Computer Engineering and Computer
Science
Department,
J.B.
Speed School of
Engineering, University of Louisville, Louisville, KY 40292


The recent dev
elopment of the Graphics Processing Unit (GPU) technology allows the
GPU to be used as a high
-
performance computing device. The computing power and
programmability of GPUs can be applied to general
-
purpose applications, such as data
processing, numerical
algorithms, and massive computation problems. The NVIDIA
Compute Unified Device Architecture (CUDA) and the Open Computing Language
(OpenCL) can be used to program these applications. CUDA is tailored towards the
NVIDIA GPUs, while OpenCL is an open stan
dard that can be designed to program not
only GPUs, but also multi
-
core chipsets from any manufacturers. In the present study,
the performance of OpenCL is compared to the performance of CUDA on data of various
sizes. The matrix arithmetic, sorting, and
other parallel algorithms are tested on CUDA
and OpenCL. The test system has an AMD triple
-
core 2.8GHz CPU and the NVIDIA
Tesla C1060. The tests measure and compare the processing time of both CUDA and
OpenCL, including data transfer time (CPU to GPU and

GPU to CPU), kernel execution
time, and the entire application running time.