User Interaction Aware Reinforcement Learning for Power and Thermal Efficiency of CPU-GPU Mobile MPSoCs

Abstract

Mobile user's usage behaviour changes throughout the day and the desirable Quality of Service (QoS) could thus change for each session. In this paper, we propose a QoS aware agent to monitor mobile user's usage behaviour to find the target frame rate, which satisfies the desired user's QoS, and applies reinforcement learning based DVFS on a CPU-GPU MPSoC to satisfy the frame rate requirement. Experimental study on a real Exynos hardware platform shows that our proposed agent is able to achieve a maximum of 50% power saving and 29% reduction in peak temperature compared to stock Android's power saving scheme. It also outperforms the existing state-of-the-art power and thermal management scheme by 41% and 19%, respectively.

CPU-GPU-Memory DVFS for Power-Efficient MPSoC in Mobile Cyber Physical Systems

“…These heterogeneous multi-processor systems have proven to provide more benefits in terms of area and core-to-application matching for improved performance, power and workload coverage [2,3]. On the other hand, given the fact that these mobile devices are battery-operated and that users expect such devices to be operable without the need for frequent charging, optimised power consumption on such devices is an important concern [4,5]. Furthermore, the PEs in these MPSoCs support dynamic voltage and frequency scaling (DVFS), which can be used to reduce dynamic power consumption (P ∝ V 2 f , where P represents dynamic power consumption, V represents the voltage of the CMOS and f represents the operating frequency) [5][6][7].…”

“…The maximum power consumption of the MPSoC when executing Streamcluster was 10.11 W. As shown in Figure 1, one interesting observation was that in a mixed workload application the memory can contribute to 19% of the total power consumption, which is still a significant amount, and hence, DVFS in memory plays an important role in regard to the total power consumption of the device. There has been a series of published studies on the effects of performing DVFS on CPU or GPU or memory separately or using a combination of two of these components [4,5,[13][14][15]; however, to the best of our knowledge there have not been any studies on the effects of performing DVFS on CPU, GPU and memory together in order to optimise the performance and power consumption of the execution of applications in mobile MPSoCs. Moreover, it is quite attractive to employ methods such as reinforcement learning (RL) to perform CPU/GPU/Memory DVFS since such methods could be application-agnostic.…”

Asynchronous Hybrid Deep Learning (AHDL): A Deep Learning Based Resource Mapping in DVFS Enabled Mobile MPSoCs

“…On the other hand, several studies, which are focused on extracting features from the source code of an application and This work was supported by Nosh Technologies under Grant nosh/agritech-000001 and by the U.K. Engineering and Physical Sciences Research Council (EPSRC) under Grant EP/P017487/1, Grant EP/R02572X/1, Grant EP/P016006/1 and Grant EP/V000462/1. then utilizing several machine learning models [8]- [13] such as Support Vector Machines (SVMs), Nearest Neighbor, etc. to classify different set of applications and then deciding the resources that need to be allocated to such applications.…”

IRON-MAN: An Approach to Perform Temporal Motionless Analysis of Video Using CNN in MPSoC

“…Although this study gave a novel approach on analyzing traffic using video cameras in low bandwidth network without the need to communicate the image frames over the WIFI network, the study had energy consumption and device lifespan reliability issues (shown later in the section). Due to wide consumer adoption of mobile devices utilizing multi-processor system-on-achip (MPSoC) [24]- [32], which implements several different types of processing elements (PEs) such as CPU/GPU on the platform, MPSoCs are perfect candidates for implementing computing resource demanding algorithms such as CNN based methodologies.…”