Molecular simulations are a set of methods for performing computer experiments on models of molecular systems. They act as a bridge between theoretical predictions and experimental results. The need for greater computational power grows with the complexity and size of the simulation model. Graphics processing units are increasingly being used for general-purpose computing due to their favourable ratio of computing capacity to power consumption and price. In our work, we focus on the Monte Carlo method for simulation of fluids. We have successfully adapted it for execution on graphics processing units using the CUDA platform and the energy decomposition principle. Throughout the simulation the system energy and radial distribution function are calculated. Inter-atom interactions are modelled using the Lennard-Jones potential. We have also implemented support for molecules composed of several different atoms. We have analysed the performance of our parallel implementation in comparison to a sequential implementation. We have achieved up to 172-fold speedups when using double precision for floating-point number representation and almost up to 640-fold speedups when using single precision.
|
