Thesis
Archon--a framework for dynamically-tuned CPU-GPU hybridization
Washington State University
Master of Science (MS), Washington State University
2016
Handle:
https://hdl.handle.net/2376/105633
Abstract
Graphics Processing Units (GPUs) have recently become widely used in general purpose computing, aiming for improving the performance of applications. However, this performance gain often comes with higher power consumption. In this work, we present Archon, a framework for power-aware CPU-GPU hybridization. Archon takes user's programs as input, automatically distributes the workload between the CPU and the GPU, and dynamically tunes the distribution ratio at runtime for an energy-efficient execution. To evaluate the effectiveness of Archon, experiments have been carried out using a variety of applications. Several of these experiments involve computer vision algorithms, which often perform reasonably well on both the CPU and the GPU. We have also evaluated Archon with matrix multiplication, as a simpler, computationally-expensive example outside the field of computer vision. The results of these experiments show us that, in many cases, Archon can achieve substantial improvements in both performance and energy consumption, with little extra effort from client programmers.
Metrics
4 File views/ downloads
19 Record Views
Details
- Title
- Archon--a framework for dynamically-tuned CPU-GPU hybridization
- Creators
- Kyle Ryan Siehl
- Contributors
- Xinghui Zhao (Degree Supervisor)
- Awarding Institution
- Washington State University
- Academic Unit
- Electrical Engineering and Computer Science, School of
- Theses and Dissertations
- Master of Science (MS), Washington State University
- Publisher
- Washington State University; [Pullman, Washington] :
- Identifiers
- 99900525380401842
- Language
- English
- Resource Type
- Thesis