An Implementation of the Smooth Particle Mesh Ewald Method on GPU Hardware.

Area of Science:

• Computational chemistry
• Molecular dynamics simulations
• High-performance computing

Background:

• The smooth particle mesh Ewald summation (SPME) method is crucial for calculating electrostatic forces in molecular dynamics (MD).
• Optimizing SPME for parallel architectures is essential for efficient simulations.
• Graphical Processing Units (GPUs) offer significant parallelism and computational power.

Purpose of the Study:

• To implement and evaluate the SPME method on Nvidia GPUs.
• To assess the performance and accuracy of the GPU-accelerated SPME implementation.
• To enable faster and more efficient biomolecular simulations.

Main Methods:

• Developed an implementation of the SPME method tailored for Nvidia GPUs.
• Integrated the GPU implementation with the ACEMD molecular dynamics package.
• Compared simulation performance and accuracy against a double-precision CPU implementation.

Main Results:

• A single GPU workstation achieved simulation rates of approximately 50 ns/day for typical biomolecular systems (e.g., DHFR, 26K atoms).
• The GPU implementation demonstrated accuracy comparable to a reference double-precision CPU implementation.
• The high parallelism and arithmetic performance of GPUs were leveraged for computational acceleration.

Conclusions:

• GPU acceleration of the SPME method significantly enhances molecular dynamics simulation efficiency.
• The developed implementation provides a viable and accurate alternative to traditional CPU-based methods.
• This work facilitates faster and more extensive investigations in computational biomolecular science.