CUDA-Accelerated Simulation of Brownian Dynamics
Kevin Dice, Cortez Gray, George Walker, Cody Baldwin, Daniel Andresen · 2018
Molecular simulations are increasingly used to study nanoscale biological and synthetic systems. To model interactions between molecules, we look to Brownian Dynamics. An n-body problem, the calculation of pairwise forces between molecules is naively an O(n2) algorithm. In our work, we begin with a single-threaded C++ code, simulating toluene (C6H5CH3) molecules suspended in water and their interaction with a graphene sheet. Our goal was to study algorithmic improvements for n-body problems and benchmark implementations along the way. We reduce computational cost with Verlet pairlists and cell decomposition and explored strategies to optimize cache utilization. Finally, we implement and benchmark a CUDA-accelerated code targeting the NVIDIA GeForce GTX 1080 Ti GPU.