Energy Efficient Continuous Multimedia Processing Using

Energy consumption is an important issue for mobile de-vices, as the technological development in battery technol-ogy has not kept pace with the power requirements of mo-bile hardware. This has led to innovative architectures with new power-saving capabilities, such as the Tegra K1 System-on-Chip (SoC). With a quad-core CPU, 192 CUDA cores and frequency scaling, the challenge in developing software for this architecture lies in optimising the energy consump-tion of mobile workloads while meeting performance require-ments. In this paper, we consider the energy consumption of continuous multimedia workloads that are performed on video streams, such as image rotation. We take an empiri-cal approach to find the most energy-efficient hardware and software configurations on a set of test videos. We find that the processor frequency should be minimised such that ap-plication deadlines are met. Using this heuristic, a reduction in energy consumption by up to 28 % can be achieved com-pared to the standard Linux frequency scaling algorithms