EFFECTIVE GPU ACCELERATION OF LARGE SCALE, ASYNCHRONOUS SIMULATIONS ON GRAPHS

Abstract

The recent emergence of GPGPU programming has resulted in a number of very efficient, but ultimately ad-hoc implementations of GPU accelerated simulations of complex systems. Because developing applications for the GPU is still a difficult and time consuming task, efficient GPU parallelizations of general purpose modeling frameworks are very useful. The dimer automaton is a stochastic modeling and simulation framework with a good balance of robustness, generality, and simplicity with capacity to model a wide range of phenomena. A major advantage of dimer automata is the ease in which they can be applied to any space that can be represented as a graph. Therefore, we have developed an efficient GPU implementation of dimer automata that runs up to 80 times faster than the serial implementation.