Abstract:
Multi-core System On Chips (SoCs) are playing a predominating role in smart phones and tablets. Unlike traditional multi-core desktops, mobile devices are restricted by t...Show MoreMetadata
Abstract:
Multi-core System On Chips (SoCs) are playing a predominating role in smart phones and tablets. Unlike traditional multi-core desktops, mobile devices are restricted by the limited energy the batteries can buffer. It remains unclear whether it is energy efficient to adopt heterogeneous computing in multi-core mobile devices. In this paper, we evaluate mobile heterogeneous computing by benchmarking RenderScript, a high performance computing framework in Android system, using 6 selected benchmarks in the area of linear algebra, machine learning and image processing. For each benchmark, both the original version and the RenderScript heterogeneous version are implemented for the performance comparison. We make a thorough study of the performance in terms of computation speedup and power consumption on two smart phone platforms. Our results demonstrate that, compared with original implementation, the increase of computation speed ranges from 2 to 18 by using RenderScript while the power consumption overhead is capped by 75%. By adopting heterogeneous computing, the total energy required for executing the applications can be significantly reduced. Based on the benchmarking results, the potential for improving energy efficiency of multi-core mobile devices is discussed.
Published in: 2015 12th Annual IEEE International Conference on Sensing, Communication, and Networking - Workshops (SECON Workshops)
Date of Conference: 22-25 June 2015
Date Added to IEEE Xplore: 16 November 2015
ISBN Information: