Molecular Dynamics in the Force Field FF14SB in Water TIP4P-Ew, and in the Force Field FF15IPQ in Water SPC/Eb: a Comparative Analysis on GPU and CPU

Abstract

A comparative analysis of computational efficiency and scalability of molecular dynamics (MD) implemented in the AMBER package was carried out on real biological systems using the classical force field FF14SB with the 4-site water model TIP4P-Ew, as well as the new promising force field FF15IPQ with the 3-site water model SPC/Eb. The Intel Xeon E5-2697 v3 processors, as well as GPU accelerators Tesla K40 (Kepler architecture) and P100 (Pascal) were used. Reduction of the number of atoms in a cell by 25–31 % as a result of implementing a 3-site solvent model speeds up the MD calculations by up to 63 % and decreases scalability by about 11 %. The obtained results can be qualitatively different, what indicates the need for joint use of different force fields at studying biological systems. The use of GPU-accelerators as an alternative to classical CPUs provides an opportunity to significantly increase the length on MD trajectories in the daily laboratory practice.