Large-Scale Molecular Dynamics Simulation Based on Heterogeneous Many-Core Architecture.

Area of Science:

• Computational chemistry
• High-performance computing

Background:

• Molecular dynamics (MD) simulations are crucial for understanding molecular behavior.
• Increasing demand for large-scale simulations and efficient task management necessitates advanced computational methods.

Purpose of the Study:

• To develop a parallel acceleration method for large-scale MD simulations on Sunway heterogeneous many-core processors.
• To address challenges in handling extensive simulation systems and numerous concurrent tasks.

Main Methods:

• Integration of task scheduling, simulation calculations, and data storage.
• Porting the GROMACS software to the Sunway architecture.
• Heterogeneous processor acceleration for short-range force calculations.

Main Results:

• Achieved approximately 10-fold acceleration and 90% scalability for single simulation tasks.
• Enabled parallel execution of numerous tasks with 90% scalability.
• Completed 50 ns simulations for over 3000 conotoxin structures in 5 hours.
• Exceeded the efficiency of GPU clusters for protein-ligand complex evaluations.

Conclusions:

• The proposed parallel acceleration method effectively enhances large-scale MD simulations on Sunway processors.
• The strategy demonstrates significant improvements in both single-task performance and multi-task handling.
• This approach offers a powerful solution for accelerating complex molecular simulations in computational chemistry.