Impact of memory frequency scaling on user-centric smartphone workloads

Abstract

Improving battery life in mobile phones has become a top concern with the increase in memory and computing requirements of applications with tough quality-of-service needs. Many energy-efficient mobile solutions vary the CPU and GPU voltage/frequency to save power consumption. However, energy-aware control over the memory bus connecting the various on-chip subsystems has had much less interest. This measurement-based study first analyse the CPU, GPU and memory cost (i.e. product of utilisation and frequency) of user-centric smartphone workloads. The impact of memory frequency scaling on power consumption and qualityof-service is also measured. We also present a preliminary analysis into the frequency levels selected by the different default governors of the CPU/GPU/memory components. We show that an interdependency exists between the CPU and memory governors and that it may cause unnecessary increase in power consumption, due to interference with the CPU frequency governor. The observations made in this measurement-based study can also reveal some design insights to system designers.

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

“…Power-saving mechanisms within performance constraints utilizing DVFS capabilities on heterogeneous MPSoCs have been considered in many studies [4][5][6]13,14,[24][25][26][27][28][29][30][31][32][33][34][35][36]. Given the fact that power consumption in a heterogeneous MPSoC can be significantly affected by big CPUs, LITTLE CPUs, GPUs and memory, most of the published studies have proposed power-saving approaches utilizing DVFS of different aforementioned components of the MPSoC but have not considered performing DVFS on all these components to achieve more reduced power consumption while catering to performance constraints.…”

“…Additionally, there have been extensive studies [31][32][33], in which DVFS was performed on memory to improve power efficiency either in general-purpose computers or server systems. Only a handful of studies [34][35][36] have performed DVFS on CPU and memory together to benefit from combined power efficiency in a mobile platform. However, none of these studies attempted to combine the benefits of performing DVFS on CPUs, GPUs and memory in conjunction in a mobile MPSoC to improve power efficiency while catering for performance constraints and hence, this paper addresses this gap in the literature.…”