NVIDIA ALCHEMI Toolkit-Ops#

GPU-accelerated computational kernels for molecular dynamics simulations and atomic-scale modeling. Built on NVIDIA Warp, this toolkit provides high-performance primitives for neighbor list construction, dispersion corrections, and other operations critical to atomistic workflows.

Key Capabilities#

  • O(N) cell list algorithms for neighbor list construction

  • DFT-D3(BJ) dispersion corrections with environment-dependent C6 coefficients

  • Ewald and particle mesh Ewald (PME) methods for electrostatic calculations

  • Batch processing for multiple systems with heterogeneous parameters

  • Native PyTorch tensor support with torch.compile compatibility

  • Dense or sparse COO output formats for graph neural networks

Who Is This For?#

ML Researchers : Integrate high-performance neighbor lists and energy corrections into graph neural network pipelines.

Method Developers : Access low-level Warp kernels to build custom atomistic workflows.

Computational Chemists : Add GPU-accelerated dispersion corrections to DFT calculations.

Get started →

User Guide#

Examples#

Benchmarks#

API#