Multi-objective Optimization on DVFS based Hybrid Systems
Mohamed Gadou, Sankeerth Reddy Mogili, Tania Banerjee, Sanjay Ranka · 2019
Energy efficiency and power minimization have become critical for high performance computing systems. Most modern processors and co-processors are equipped with dynamic voltage and frequency scaling (DVFS) mechanisms where the operating frequency of a processor can be changed to lower power consumption. Additionally, only a subset of processors can be used to save overall energy when the scaling of the application does match with the increase in power requirements. In this paper, we present a systematic approach for deriving energy performance trade-offs on a hybrid multicore (CPU+GPU) processor. Using a proxy application for compressible multiphase turbulence that is representative of a large number of applications, we show how to derive energy performance trade-offs on a server consisting of multicore and multiple GPU processors.