Energy-efficient global illumination algorithms for mobile devices using dynamic voltage and frequency scaling
- 주제(키워드) DVFS , Path tracing , Global illumination , OpenCL , GPGPU , Mobile device
- 주제(기타) Computer Science, Software Engineering
- 설명문(일반) [Kim, SeongKi] Keimyung Univ, 1095 Dalgubeol Daero, Daegu, South Korea; [Harada, Takahiro] Adv Micro Devices Inc, One AMD Pl, Sunnyvale, CA 94085 USA; [Kim, Young J.] Ewha Womans Univ, Dept Comp Sci & Engn, 52 Ewhayeodae Gil, Seoul, South Korea
- 관리정보기술 faculty
- 등재 SCIE, SCOPUS
- 발행기관 PERGAMON-ELSEVIER SCIENCE LTD
- 발행년도 2018
- 회의명 Computer-Aided Design and Computer Graphics (CAD/GRAPHICS) Conference
- 개최지 Changsha, PEOPLES R CHINA
- 일자 AUG 26-27, 2017
- URI http://www.dcollection.net/handler/ewha/000000150111
- 본문언어 영어
- Published As http://dx.doi.org/10.1016/j.cag.2017.07.017
초록/요약
With increases of display resolution and graphical quality demands, global illumination techniques are sought after to meet such demands. However, global illumination algorithms have a long processing time on mobile devices, such as smartphones or tablet PCs. Besides the performance issue, the algorithms consume a large amount of battery-powered energy. The performance and energy consumption have a trade-off relationship. The dynamic voltage and frequency scaling (DVFS) algorithms control the balance between the performance and the energy consumption by adjusting the GPU and CPU's frequencies. To improve the performance of the global illumination algorithm and reduce the high energy consumption of the current DVFS algorithm, we suggest new algorithms with new metrics for estimating the amount of workload for GPUs as well as their memory workload. Using our new DVFS algorithms, we increase the energy efficiency, the performance per watt, of a global illumination algorithm by 33.59% in comparison to a conventional general-purpose DVFS algorithm, without modifying the underlying algorithm core. (C) 2017 Elsevier Ltd. All rights reserved.
more