Trade-off between low power and energy efficiency in benchmarking

Abstract

Microprocessor architecture and its hardware implementation have a great influence on computer performance, power consumption, and energy efficiency. Nowadays, energy efficiency is a key point of Green Computing. However improving the performance by rising the CPU clock rate usually increases both the power and the energy consumptions. In this paper, we formulate experiments to study the relation between the CPU execution time, the power consumption, and the dissipated energy over a 0.2 GHz to 1.7 GHz range of CPU frequencies. These experiments are implemented on battery-powered mobile systems, for specific Dynamic Voltage and Frequency Scaling (DVFS) settings, by executing custom-made and standard benchmark tools for Intel and ARM processors. We report the results of the experiments and we demonstrate that we can achieve a minimum energy consumption but with a limited performance improvement. On the other hand, the performance optimization always increases the power. So other design factors, rather than clock speed, have to be considered for a low-power achievement.