Performance aspects of collocated and staggered grids for particle-in-cell plasma simulation

We present a computational comparison of collocated and staggered uniform grids for particle-in-cell plasma simulation. Both types of grids are widely used, and numerical properties of the corresponding solvers are well-studied. However, for large-scale simulations performance is also an important factor, which is the focus of this paper. We start with a baseline implementation, apply widely-used techniques for performance optimization and measure their efficacy for both grids on a high-end Xeon CPU and a second-generation Xeon Phi processor. For the optimized version the collocated grid outperforms the staggered one by about 1.5x on both Xeon and Xeon Phi. The speedup on the Xeon Phi processor compared to Xeon is about 1.9x