Tinker-HP : Accelerating Molecular Dynamics Simulations of Large Complex Systems with Advanced Point Dipole...

Area of Science:

• Computational chemistry and biophysics
• High-performance computing (HPC)

Background:

• Molecular dynamics (MD) simulations are crucial for understanding biomolecular systems.
• Polarizable many-body force fields, like AMOEBA, offer high accuracy but are computationally intensive.
• Accelerating these simulations is essential for tackling larger and more complex biological problems.

Approach:

• Extended the Tinker-HP package to leverage Graphics Processing Unit (GPU) acceleration for MD simulations.
• Developed a scalable strategy using OpenACC and CUDA for efficient single- and multi-GPU utilization.
• Implemented and optimized the AMOEBA polarizable force field for GPU architectures.
• Investigated multi-precision arithmetic, demonstrating the performance benefits of lower precision with minimal accuracy loss.

Key Points:

• Achieved significant speedups in MD simulations using GPU acceleration on various NVIDIA platforms (2080Ti, 3090, V100, A100).
• Demonstrated efficient performance for large biosystems with millions of atoms.
• Showcased the benefits of mixed-precision (lower precision arithmetic) for enhanced computational speed without compromising accuracy.
• The enhanced Tinker-HP package provides state-of-the-art performance for AMOEBA force field simulations.

Conclusions:

• The GPU-accelerated Tinker-HP package dramatically reduces simulation time for polarizable force fields.
• This advancement enables large-scale biophysics research and contributes to high-performance computing efforts.
• The software is freely available on GitHub for academic use, fostering community collaboration.