Parts | Days | Selection | Search | Updates | Downloads | Help
CPP: Fachverband Chemische Physik und Polymerphysik
CPP 20: Poster: Computational Soft Matter Physics
CPP 20.2: Poster
Tuesday, March 15, 2011, 18:00–20:00, P2
Parallel Monte Carlo simulation of lattice polymers on graphics processing units — •Marco Werner1,2 and Jens-Uwe Sommer1,2 — 1Leibniz-Institut für Polymerforschung Dresden, Germany — 2Technische Universität Dresden - Institut für Theoretische Physik
A parallel version of the bond fluctuation model (BFM) was developed for programmable graphics processing units (GPUs). The BFM is a lattice-based dynamical Monte Carlo method addressing the universal dynamic and static properties of polymer structures [I. Carmesin, K. Kremer, Macromolecules, 21, 2819, (1988)]. Recent developments by graphic card industry give rise to new possibilities in terms of system size and simulation speed accessible by the BFM. Modern GPUs provide several hundreds of parallel processors which work most efficient with massively parallel processes. We make use of this architecture by associating each monomer to one parallel process. In one elementary step all monomers are tried to move simultaneously. Hereby, the number of random memory accesses is reduced and the rate of tried moves per monomer increases with increasing number of monomers. The statics and dynamics of samples like polymer melts are compared between parallel and serial BFM-code to ensure the equivalence both methods. On the latest generation of GPU we obtain a maximum up-speed of about 50 compared to actual serial processors.