Simulation of a Granular Flow Experiment Using GPU Parallelism

We describe a computer simulation of avalanches on a conical bead pile using both central processing units (CPUs) and graphical processing units (GPUs) to increase computational speed on a single desktop computer. We make detailed comparisons with real experiments involving physical beads. In the experimental system, the avalanche size distribution follows a power law for beads dropped onto the pile apex from a low drop height. The simulation models each bead independently with its own position and velocity. As each bead is independent, we can use parallelism to thread various computations on the beads to the GPU. This allows the simulation to run significantly faster than the physical experiment while still involving tens of thousands of beads. By monitoring every bead, we can extend the research beyond what is experimentally possible. The experiments are sensitive only to avalanches that exit the pile, while the simulation is also sensitive to avalanches confined to the pile.