GPU-Accelerated Exploration of Biomolecular Energy Landscapes

Abstract

We present graphics processing unit (GPU)-acceleration of various computational energy landscape methods for biomolecular systems. Basin-hopping global optimization, the doubly nudged elastic band method (DNEB), hybrid eigenvector-following (EF), and a local rigid body framework are described, including details of GPU implementations. We analyze the results for eight different system sizes, and consider the effects of history size for minimization and local rigidification on the overall efficiency. We demonstrate improvement relative to CPU performance of up to 2 orders of magnitude for the largest systems.