high performance computing on graphics processing units: hgpu.org

Molecular dynamics simulations allow us to study the behavior of complex biomolecular systems by modeling the pairwise interaction forces between all atoms. Molecular systems are subject to slowly decaying electrostatic potentials, which turn molecular dynamics into an n-body problem. In this paper, we present a parallel and scalable solution to compute long-range molecular forces, based on the multilevel summation method (MSM). We first demonstrate an optimization of MSM that replaces 3D convolutions with FFTs, and we achieve a single-GPU performance comparable to the particle mesh Ewald (PME) method, the de facto standard for long-range molecular force computation. But most importantly, we propose a distributed MSM that avoids the scalability difficulties of PME. Our distributed solution is based on a spatial partitioning of the MSM multilevel grid, together with massively parallel algorithms for interface update and synchronization. We demonstrate the scalability of our approach on an on-board multi-GPU platform.