high performance computing on graphics processing units: hgpu.org

Computer simulations play a vital role in understanding the phase behavior of colloidal dispersions, however, most simulation results suffer from finite-size effects. These finite-size effects can be eliminated by finite-size scaling or by simulating large system sizes. In this thesis we show how to simulate large system sizes efficiently on Graphical Processing Units (GPUs). Whereas efficient GPU Molecular Dynamics implementations are readily available, Monte Carlo (MC) simulations have not yet received much attention, mainly due to their serial nature. We present a GPU implementation of the Monte Carlo simulation using the Hybrid Monte Carlo (HMC) method. Our implementations for long and short-ranged potentials are at least one order of magnitude faster than the regular CPU implementations. We show the correctness of our implementation by reproducing the equation of state for the three-dimensional Lennard-Jones system.