Parts | Days | Selection | Search | Updates | Downloads | Help

SOE: Fachverband Physik sozio-ökonomischer Systeme

SOE 5: Focus Session: GPU Computing (with DY)

SOE 5.2: Invited Talk

Monday, March 14, 2011, 14:30–15:00, GÖR 226

Accelerating Monte Carlo Simulations in Statistical Physics with GPU's — •David Landau and Junqi Yin — University of Georgia, Athens, GA 30622, U.S.A.

High resolution Monte Carlo simulations are often limited by available resources. This is because long sampling times and large systems are often needed to overcome long correlation times and finite size effects for the systems under study. We will describe how GPU's can be used to formulate multi-threaded algorithms that dramatically accelerate performance of Monte Carlo simulations of condensed matter systems. We give examples of the application to parallel tempering simulation of magnetic lattice models and Wang-Landau sampling of water clusters in the continuum. In both cases a speedup of more than a factor of $50$ was achieved compared to a single, current generation CPU; moreover, our implementation scales nearly linearly with the number of GPU's.