Machine learning method for energy reduction by utilizing dynamic mixed precision on GPU-based supercomputers

摘要

In this work,we propose a method that allows us to reduce energy consumption of an applicationexecuted on supercomputing centers. The proposed method is based on a mixed precision arithmeticwhere the precision of data is calibrated at runtime. For this reason, we develop a modifiedversion of the random forest algorithm. The effectiveness of the proposed approach is validatedwith a real-life scientific application called MPDATA, which is part of the numerical model usedin weather forecasting. The energy efficiency of the proposed method is examined using twoGPU-based clusters. The first of them is the Piz Daint supercomputer, currently ranked 3rd at theTOP500 list (November 2017). It is equippedwith NVIDIA Tesla P100GPUaccelerators based onthe Pascal architecture. The second is theMICLAB cluster containing NVIDIA Tesla K80 based onthe Kepler architecture. The achieved results show that the proposed machine learning methodallows us to provide the accuracy of computation comparable with that achieved double precisionand reduce the energy consumption up to 36% compared to the double precision version ofMPDATA.