The Heisenberg spin glass model on GPU: myths and actual facts

TL;DR

This paper explores GPU implementations of the 3D Heisenberg spin glass model, demonstrating that shared memory enhances performance when combined with a multi-hit technique, challenging some common assumptions.

Contribution

It provides a detailed comparison of GPU memory strategies for simulating the 3D Heisenberg spin glass model, highlighting the conditions under which shared memory improves performance.

Findings

Shared memory outperforms global memory with multi-hit techniques.

Performance gains depend on memory strategy and implementation details.

The study clarifies misconceptions about GPU memory usage for spin glass simulations.

Abstract

We describe different implementations of the 3D Heisenberg spin glass model for Graphics Processing Units (GPU). The results show that the {\em fast} shared memory gives better performance with respect to the {\em slow} global memory only if a multi-hit technique is used.