Empirical characterization of power efficiency for large scale data processing
Yongbin Lee, Sungchan Kim · 2015
It becomes popular to equip CPU and GPU on a single computer system because of its performance and energy benefits, constituting a heterogeneous system for processing big data workloads. However, the optimal exploitation of such a heterogeneous system requires us to know the power consumption characteristics of the applications for difference processing units. To this end, this paper aims at characterizing the power efficiency of CPUs and GPUs for big data processing through empirical measurements. We take three recent computing units, high-end CPU, and GPU, and mobile embedded GPU as target platforms. We first show the performance and power consumption measurements on each computing platform using the Rodinia benchmarks as representative big data workloads. Then, we discuss how performance-per-watt of each computing platform is associated with different characteristics of the workloads.