Algorithms for GPU-based molecular dynamics simulations of complex fluids: Applications to water, mixtures, and...

Area of Science:

• Computational Chemistry
• Materials Science
• Physics

Background:

• Molecular dynamics (MD) simulations are crucial for understanding molecular behavior.
• Traditional MD codes often face computational limitations, especially for large systems.
• Accelerating MD simulations is key to advancing scientific discovery.

Purpose of the Study:

• To develop a custom, high-performance molecular dynamics simulation code.
• To enable efficient simulations of multicomponent fluids on graphics processing units (GPUs).
• To analyze the performance gains of GPU acceleration for various systems.

Main Methods:

• Implementation of a double-precision MD code optimized for CUDA-enabled NVIDIA GPUs.
• Inclusion of algorithms for intramolecular/intermolecular forces, coarse-grained/atomistic models, constraints, thermostats, and distribution functions.
• Development of neighbor list algorithm for improved system size scaling and computation of Lennard-Jones, Gay-Berne, and electrostatic (Ewald summation) interactions.

Main Results:

• Significant performance increases were observed across three test systems: SPC/E water, n-hexane/2-propanol mixture, and a liquid crystal mesogen.
• Single GPU usage resulted in 33-119 fold performance increase compared to serial code.
• Multiple GPUs demonstrated substantial speedups: 69-287 fold with two GPUs and 101-377 fold with three GPUs.

Conclusions:

• The custom GPU-accelerated MD code offers substantial computational advantages.
• The code effectively simulates complex molecular systems with high efficiency.
• This approach paves the way for more extensive and complex molecular simulations.