An overview of techniques for predicting the performance of GPU accelerated applications
Instituto de Informatica, Universidade Federal do Rio Grande do Sul, Brazil
Universidade Federal do Rio Grande do Sul, 2012
@article{tesser2012overview,
title={An overview of techniques for predicting the performance of GPU accelerated applications},
author={Tesser, Rafael Keller and Navaux, Philippe OA},
year={2012}
}
The ability to predict the performance of applications in large-scale parallel systems is essential. One of the main incentives for this is the high cost of executing non-production tasks on these systems. An entity may also want to predict the performance in a system that does not yet exist. One popular alternative for increasing a systems performance is the use of accelerators. This created the need for new methods for performance prediction, that take into account the use of these components. In this paper, we will present some recent works in the area of performance prediction in systems with accelerators. More specifically, we will focus on systems that use Graphics Processing Units (GPUs) as accelerators.
July 22, 2013 by hgpu
