6108

Improving Energy Efficiency of GPU based General-Purpose Scientific Computing through Automated Selection of Near Optimal Configurations

Xiaohan Ma, Marion Rincon, Zhigang Deng
Computer Science Department, University of Houston, Houston, TX, 77204, USA
Department of Computer Science, University of Houston, Technical report UH-CS-11-08, 2011

@article{ma2011improving,

   title={Improving Energy Efficiency of GPU based General-Purpose Scientific Computing through Automated Selection of Near Optimal Configurations},

   author={Ma, X. and Rincon, M. and Deng, Z.},

   year={2011}

}

Download Download (PDF)   View View   Source Source   

737

views

Modern GPUs have been rapidly and increasingly used as a powerful engine for a variety of general-purpose computing applications due to their enormous parallelism and throughput capabilities. However, GPU power consumption still remains high since more and more transistors are integrated into its chip. Until now, how to increase and optimize energy efficiency (e.g., performance-per-Watt ratio) of GPU-based computing applications is still a largely unsolved challenge. In this paper, we propose a novel framework to improve the energy efficiency of GPU-based General-Purpose Computing (GPGPU) applications. Based on a statistical regression model capable of dynamically estimating the runtime GPU power consumption, our framework can infer and select near-optimal GPGPU programming configurations to improve the energy efficiency of any given GPGPU program. Through preliminary empirical validation of a number of GPGPU benchmarks, we demonstrated that our framework can be robustly used to measurably improve the energy efficiency of various GPGPU programs.
No votes yet.
Please wait...

* * *

* * *

HGPU group © 2010-2017 hgpu.org

All rights belong to the respective authors

Contact us: