We study a cellular automaton derived from the phenomenon of magnetic flux creep in two-dimensional granular superconductors. We model the superconductor as an array of Josephson junctions evolving according to a set of coupled ordinary differential equations. In the limit of slowly increasing magnetic field, we reduce these equations to a simple cellular automaton. The resulting discrete dynamics, a stylized version of the continuous dynamics of the differential equations, is equivalent to the dynamics of a gradient sand pile automaton. We study the dynamics as we vary the symmetry of the underlying lattice and the shape of its boundary. We find that the "simplest" realization of the automaton, on a square lattice with commensurate boundaries, results in especially simple dynamics, while "generic" realizations exhibit more complicated dynamics characterized by statistics with broad distributions, even in the absence of noise or disorder. This research was partially supported by NSF-DMR 9987850.