The rise of machine learning workload on smart-phones has propelled GPUs into one of the most power hungry components of modern smartphones. Optimizing the power consumption of mobile GPUs in turn requires accurate estimation of their power draw during app execution. We observe that the prior-art, utilization-frequency based GPU models cannot capture the diverse micro-architectural usage of modern mobile GPUs, and study whether performance monitoring counter (PMC)-based models recently proposed for desktop/server GPUs can be applied to accurately model mobile GPU power. Our study shows that the PMCs that come with dominating mobile GPUs used in modern smartphones are sufficient to model mobile GPU power, but exhibit multicollinearity if used altogether. We present APGPM, the first mobile GPU power modeling methodology that automatically selects an optimal set of PMCs that maximizes the GPU power model accuracy. Evaluation on the two representative mobile GPUs shows that APGPM-generated GPU power models reduce the MAPE modeling error of prior-art by 11.3% to 15.4% while using only 4.66% to 20.41 % of the total number of available PMCs.

By omitting external lighting, OLED display significantly reduces the power draw compared to its predecessor LCD and has gained wide adoption in modern smartphones. The real potential of OLED in saving phone battery drain lies in exploiting app UI color design, i.e., how to design app UI to use pixel colors that result in low OLED display power draw. In this paper, we design and implement an accurate per-frame OLED display power profiler, PFOP, that helps developers to gain insight into the impact of different app UI design on its OLED power draw, and an enhanced Android Battery that helps phone users to understand and manage phone display energy drain, for example, from different app and display configurations such as dark mode and screen brightness. A major challenge in designing both tools is to develop an accurate and robust OLED display power model. We experimentally show that linear-regression-based OLED power models developed in the past decade cannot capture the unique behavior of OLED display hardware in modern smartphones which have a large color space and propose a new piecewise power model that achieves much better modeling accuracy than the prior-art by applying linear regression in each small regions of the vast color space. Using the two tools, we performed to our knowledge the first power saving measurement of the emerging dark mode for a set of popular Google Android apps.

With the advent of bandwidth-hungry applications and the new advancements in wireless LAN standards (802.11ad, 802.11ay) and cellular technologies (5G), modern smartphones need to be able support multi-Gbps data rates. In this work, we explore if today's smartphones are capable of handling such high-speed network traffic. Using two high-end smartphones, we show that, contrary to previous beliefs, they can indeed support Gbps data rates without significant strain on their hardware resources. Using projections, we further show that up to 12.8 Gbps could be supported with just 50% CPU utilization. Finally, we explore the factors that make this possible and the contribution of each of them.