(214v) GPU Accelerated Configurational Bias Monte Carlo Simulations of Branched Molecules
AIChE Annual Meeting
2013
2013 AIChE Annual Meeting
Computational Molecular Science and Engineering Forum
Poster Session: Computational Molecular Science and Engineering Forum (CoMSEF)
Monday, November 4, 2013 - 6:00pm to 8:00pm
The simulation of systems containing over 100,000 atoms has become commonplace with the development open-source parallelized molecular dynamics simulation codes, such as LAMMPS and NAMD [1, 2]. In addition to outstanding scaling across CPUs, these codes, and others, such as HOOMD [3], have been written to utilize inexpensive graphics processors. For the simulation of processes at constant temperature and or pressure, molecular dynamics is clearly the method of choice, however, chemical or biological processes that occur at constant chemical potential require simulation in an ensemble that allows for fluctuations in the number of molecules. Simulation in open systems is performed typically with Monte Carlo simulations in either the grand canonical or Gibbs ensembles. However, the large system sizes [4, 5] required for biomolecular systems make such calculations computationally prohibitive using existing serial codes.
In this work, a multi-ensemble Monte Carlo simulation engine, GO-MC(GPU-Optimized Monte Carlo), capable of simulating systems of tens of thousands of molecules of arbitrary geometry on a desktop workstation is presented. The engine is built on past code to simulate linear molecules on the GPU [6, 7] and is written in C++ using NVIDIA Corp.’s Compute Unified Device Architecture (CUDA) API. Performance comparisons are made between GPU implementations of the coupled-decoupled and a coupled configurational bias methods [8-10]; as measured by acceptance rates and computation time on the device. The coupled approach, while inefficient on the CPU, shows promise on the GPU due to better thread balancing. Performance improvements due to the use of a nearest neighbor cell list [11] are discussed. Further code optimizations available on the NVIDIA Kepler architecture are described.
1. Brown, W.M., et al., Implementing molecular dynamics on hybrid high performance computers - short range forces. Computer Physics Communications, 2011. 182(4): p. 898-911.
2. Phillips, J.C., J.E. Stone, and K. Schulten, Adapting a message-driven parallel application to GPU-accelerated clusters, in Proceedings of the 2008 ACM/IEEE conference on Supercomputing2008, IEEE Press: Austin, Texas. p. 1-9.
3. Nguyen, T.D., et al., Rigid body constraints realized in massively-parallel molecular dynamics on graphics processing units. Computer Physics Communications, 2011. 182(11): p. 2307-2313.
4. Freddolino, P.L., et al., Molecular Dynamics Simulations of the Complete Satellite Tobacco Mosaic Virus. Structure (London, England : 1993), 2006. 14(3): p. 437-449.
5. Sanbonmatsu, K.Y., S. Joseph, and C.-S. Tung, Simulating movement of tRNA into the ribosome during decoding. Proceedings of the National Academy of Sciences of the United States of America, 2005. 102(44): p. 15854-15859.
6. Jason R. Mick, K.R., Eyad Hailat, Vincent Russo, Loren Schwiebert, Jeffrey J. Potoff, GPU Accelerated Configurational Bias Monte Carlo Simulations of Linear Alkanes. Proceedings of the 2013 AIChE Annual Meeting, 2012: p. 51e.
7. Jason R. Mick, K.R., Eyad Hailat, Vincent Russo, Loren Schwiebert, Jeffrey J. Potoff, GPU-accelerated Gibbs ensemble Monte Carlo simulations of Lennard-Jonesium” Comp. Phys. Comm. (submitted, in review), 2013.
8. Frenkel, D. and B. Smit, Understanding molecular simulation: from algorithms to applications2002: Academic Press.
9. Martin, M.G. and J.I. Siepmann, Novel Configurational-Bias Monte Carlo Method for Branched Molecules. Transferable Potentials for Phase Equilibria. 2. United-Atom Description of Branched Alkanes. The Journal of Physical Chemistry B, 1999. 103(21): p. 4508-4517.
10. Laso, M., J.J.d. Pablo, and U.W. Suter, Simulation of phase equilibria for chain molecules. The Journal of Chemical Physics, 1992. 97(4): p. 2817-2819.
11. Mazzeo, M.D., M. Ricci, and C. Zannoni, The Linked Neighbour List (LNL) method for fast off-lattice Monte Carlo simulations of fluids. Computer Physics Communications, 2010. 181(3): p. 569-581.