Routine Microsecond Molecular Dynamics Simulations with AMBER on GPUs. 2. Explicit Solvent Particle Mesh Ewald

Area of Science:

• Computational Chemistry
• Molecular Dynamics Simulations
• GPU Computing

Background:

• Classical molecular dynamics (MD) simulations are crucial for understanding molecular behavior.
• Traditional CPU-based simulations face performance limitations for large systems.
• Accelerating MD simulations on Graphics Processing Units (GPUs) offers a promising solution.

Purpose of the Study:

• To present an optimized implementation of explicit solvent all-atom classical MD within the AMBER package for CUDA-enabled GPUs.
• To validate the performance and accuracy of the GPU implementation against CPU-based methods.
• To introduce and evaluate novel precision models for GPU-accelerated MD.

Main Methods:

• Implementation of MD simulations on CUDA-enabled NVIDIA GPUs using the AMBER package.
• Support for NVE, NVT, and NPT statistical mechanical ensembles.
• Utilized Particle Mesh Ewald (PME) for long-range electrostatics.
• Developed and tested three precision models: SPDP, SPFP, and DPDP.

Main Results:

• The GPU implementation achieves statistically indistinguishable results from the CPU version.
• Performance significantly exceeds that of traditional CPU-based clusters and supercomputers.
• Demonstrated validation of the implementation and precision models, highlighting substantial performance improvements.

Conclusions:

• The GPU-accelerated AMBER package provides a powerful and efficient tool for molecular dynamics simulations.
• The new precision models offer viable alternatives for GPU-based MD, with performance benefits.
• The study validates the effectiveness of GPU computing for advancing molecular dynamics research.