Power Measurements of Hartree-Fock Algorithms Using Different Storage Devices
Vladimir Mironov, Alexander A. Moskovsky, Yuri Alexeev · 2017
In this paper, we analyze energy consumption characteristics of calculation runs with different implementations of the fundamental quantum chemistry method called Hartree-Fock in the popular quantum chemistry package GAMESS. The Hartree-Fock algorithms vary in how electron repulsion integrals are recomputed or stored in memory or disk. Based on our results, storing electron repulsion integrals in memory is shown to be more energy efficient if the storage bandwidth is equal to or higher than the speed of integral generation. Our theoretical models predict that the optimal strategy for improving energy efficiency for Hartree-Fock is to re-compute integrals with high angular momentum and store other integrals in memory or with the recent generation of solid state drives (SSDs). According to our results, the overall energy saving from storing low angular momentum integral in memory rather than re-computing them is 6-10 times for modern Intel® processors. It is, to the best of our knowledge, the first study to improve energy efficiency of Hartree-Fock by examining different options for storing various types of electron repulsion integrals.