Graphics Processing Unit acceleration of the Random Phase Approximation in the projector augmented wave method

TL;DR

This paper demonstrates GPU acceleration of the Random Phase Approximation in the projector augmented wave method, significantly reducing computation time and enabling routine surface chemistry simulations.

Contribution

The implementation of GPU acceleration for RPA in the gpaw code transforms the computation from memory-bound to compute-bound, achieving substantial speedups.

Findings

Speedups of 10+ to 40+ times with 8 GPUs versus 8 CPUs

A realistic RPA calculation for CO/Ni(111) completed in 5.5 hours using 8 GPUs

GPU acceleration makes routine surface chemistry RPA calculations feasible

Abstract

The Random Phase Approximation (RPA) for correlation energy in the grid-based projector augmented wave (gpaw) code is accelerated by porting to the Graphics Processing Unit (GPU) architecture. The acceleration is achieved by grouping independent vectors/matrices and transforming the implementation from being memory bound to being computation/latency bound. With this approach, both the CPU and GPU implementations have been enhanced. We tested the GPU implementation on a few representative systems: molecules (O2), bulk solids (Li2O and MoO3) and molecules adsorbed on metal surfaces (N2/Ru(0001) and CO/Ni(111)). Improvements from 10+ to 40+ have been achieved (8-GPUs versus 8-CPUs). A realistic RPA calculation for CO/Ni(111) surface can be finished in 5.5 h using 8 GPUs. It is thus promising to employ non-self-consistent RPA for routine surface chemistry simulations.