News

NVIDIA Launches cuPotential to Accelerate Molecular Dynamics Simulations

NVIDIA introduces cuPotential, a library to optimize molecular dynamics simulations with modular C API primitives and OpenMM integration.

NVIDIA Launches cuPotential to Accelerate Molecular Dynamics Simulations

NVIDIA has unveiled cuPotential, a specialized library designed to accelerate molecular dynamics (MD) simulations by addressing computational bottlenecks in evaluating long-range interactions under periodic boundary conditions. The library introduces modular primitives with a C API, enabling developers to optimize workloads across heterogeneous GPU hardware. This release also includes a plugin for OpenMM, a widely used MD engine, allowing seamless integration with minimal code modifications.

Key Features of cuPotential

cuPotential provides utilities for both direct and reciprocal space calculations, including cutoff-based interaction evaluators and charge spreading functions. The library incorporates advanced splitting and spreading methods developed in collaboration with D. E. Shaw Research and the Flatiron Institute. These methods achieve high accuracy at reduced grid resolutions and cutoffs, improving computational efficiency without compromising simulation fidelity.

Benchmarks conducted on GPU architectures ranging from NVIDIA T4 to RTX PRO 6000 Blackwell demonstrate end-to-end speedups of 2x to over 3x for systems containing 24,000 to 400,000 atoms under NPT conditions. Additionally, energy drift measurements over 10 nanoseconds show that cuPotential reduces drift by several hundredfold compared to default Particle Mesh Ewald (PME) settings, exceeding typical biomolecular simulation requirements.

Integration with OpenMM

The cuPotential OpenMM plugin replaces the default nonbonded force, Langevin integrator, and Monte Carlo barostat in OpenMM simulations. Developers can integrate the plugin by substituting the system's nonbonded force and using cuPotential's integrator and barostat. This drop-in approach simplifies adoption while maintaining compatibility with existing workflows.

To build and install the plugin, users need a Linux x86-64 system with a supported NVIDIA GPU, a compatible NVIDIA driver, and Conda or Micromamba for environment management. The installation process involves creating a Conda environment, configuring build helpers, and running the provided molecular dynamics samples to verify functionality. The samples evaluate systems such as DHFR, BACE, and ApoA1 under NVT and NPT conditions, measuring both accuracy and performance.

Scientific Advancements

cuPotential enhances the accuracy of electrostatics and dispersion energy calculations while maintaining comparable force errors at lower computational costs. For electrostatics, it uses a double-precision reference with optimized cutoff lengths and mesh resolutions. Dispersion energy accuracy improves by several thousandfold, ensuring reliable results for biomolecular and materials science simulations.

The library's rigorous attention to accuracy and energy drift allows researchers to leverage GPU acceleration without compromising the integrity of their simulations. By addressing key computational challenges, cuPotential represents a significant advancement in the field of molecular dynamics.

Sign in to save this article Free account: build your research collection and return to read it anytime.