Fraunhofer-Gesellschaft

Publica

Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

GPU metrics for a linear solver

 
: Panchatcharam, M.; Sundar, S.; Klar, A.

Neural, Parallel and Scientific Computations 21 (2013), No.3-4, pp.361-374
ISSN: 1061-5369
English
Journal Article
Fraunhofer ITWM ()

Abstract
GPU metrics is a set of measurements used to distinguish an efficient GPU and GPU algorithms. In this paper, we present a mathematical definition of three important GPU metrics such as occupancy, average bandwidth utilization and volume in order to prove certain preliminary results. We discuss another most important and frequently used GPU metric called acceleration ratio. Finally, we discuss the importance and application of GPU metrics on BICGStab algorithm.

: http://publica.fraunhofer.de/documents/N-350795.html