Boosting Free-Energy Perturbation Calculations with GPU-Accelerated NAMD

• DOI: 10.1021/acs.jcim.0c00745
• Publication Date: 2020-08-17T21:48:27Z
• Authors: Haochuan Chen; Julio D. C. Maia; Brian K. Radak; David J. Hardy; Wensheng Cai; Christophe Chipot; Emad Tajkhorshid
• Keywords: Free energy perturbation; Code (set theory); Double-precision floating-point format

Harnessing the power of graphics processing units (GPUs) to accelerate molecular dynamics (MD) simulations in context free-energy calculations has been a longstanding effort toward development versatile, high-performance MD engines. We report new GPU-based implementation NAMD perturbation (FEP), one oldest, most popular importance-sampling approaches for determination differences that underlie alchemical transformations. Compared CPU available since 2001 NAMD, our benchmarks indicate FEP traditional GPU code is about four times faster, without any noticeable loss accuracy, thereby paving way more affordable on large biological objects. Moreover, we have extended this path highly optimized single-GPU node, which proves be up nearly 30 faster than implementation. Through performance, present developments provide community with cost-effective solution conducting calculations. The FEP-enabled released 3.0.