Performance potential for simulating spin models on GPU

TL;DR

This paper evaluates the performance of GPU-based simulations of classical spin models, demonstrating that tailored algorithms can achieve speed-ups of up to 1000 times over CPU implementations.

Contribution

It shows how specific algorithm optimizations enable GPUs to realize their theoretical performance potential in spin model simulations.

Findings

Speed-ups of up to 1000x over CPU code

Effective algorithm tailoring for GPU architectures

Successful simulation of various spin models

Abstract

Graphics processing units (GPUs) are recently being used to an increasing degree for general computational purposes. This development is motivated by their theoretical peak performance, which significantly exceeds that of broadly available CPUs. For practical purposes, however, it is far from clear how much of this theoretical performance can be realized in actual scientific applications. As is discussed here for the case of studying classical spin models of statistical mechanics by Monte Carlo simulations, only an explicit tailoring of the involved algorithms to the specific architecture under consideration allows to harvest the computational power of GPU systems. A number of examples, ranging from Metropolis simulations of ferromagnetic Ising models, over continuous Heisenberg and disordered spin-glass systems to parallel-tempering simulations are discussed. Significant speed-ups by factors of up to 1000 compared to serial CPU code as well as previous GPU implementations are observed.