---
id: nvalchemi-toolkit-ops
version: "0.4.1"
license: Apache-2.0
license_treatment: permissive
maintenance: active
---

# nvalchemi-toolkit-ops — High-performance NVIDIA Warp primitives for GPU-enabled computational chemistry and atomistic simulation workflows.

License: permissive · Maintenance: active · Downloads: 238.4K/mo

## What it is and what it does

NVIDIA ALCHEMI Toolkit-Ops is a collection of GPU-optimized kernels written in warp-lang for accelerating atomistic simulations. It provides batched, high-throughput primitives for neighbor list computation (naive, cell-list, and tiled cluster-pair methods), molecular dynamics (NVE, NVT, NPT, NPH ensembles with multiple thermostat options), geometry optimization (FIRE and FIRE2), and interatomic interactions including DFT-D3 dispersion and electrostatics (DSF, Ewald, PME). The package targets systems with large atom counts and microsecond-scale per-atom throughput on GPUs.

It integrates with PyTorch and JAX, enabling differentiable computation of forces, charge gradients, virials, and stress tensors. The kernels are modular and reusable, intended for library developers, researchers developing new methods, and engineers building production molecular dynamics or interatomic potential applications. Installation is straightforward (pure Python wheel with numpy and warp-lang dependencies), but requires PyTorch or JAX with CUDA support and Python 3.11–3.14.

Use it for:

- Accelerate existing molecular dynamics workflows by replacing CPU neighbor list computation with GPU-batched cell-list or cluster-pair methods.
- Compute DFT-D3 dispersion corrections on batches of molecules during model training with automatic differentiation.
- Build production molecular dynamics simulations with NVT or NPT ensembles using GPU-optimized Langevin or Nosé-Hoover thermostats.
- Perform geometry optimization (lattice and coordinate relaxation) via FIRE2 on large systems without implementing custom GPU kernels.
- Evaluate particle mesh Ewald electrostatics with automatic parameter tuning and backpropagation for charge-aware training.

## Worth the install?

AI-flagged interpretation of the facts above — verify before relying.

GPU-accelerated batched primitives for atomistic simulation: neighbor lists, molecular dynamics ensembles, geometry optimization, and interatomic interactions (dispersion, electrostatics) built on NVIDIA Warp with PyTorch and JAX bindings.

Yes, if you are working on GPU-accelerated atomistic simulations or molecular dynamics and need production-ready, batched kernels. The package is actively maintained, has no known vulnerabilities, carries a permissive Apache-2.0 license, and integrates cleanly with PyTorch and JAX. Install friction is low. The main constraint is the requirement for CUDA 12 or 13 and Python 3.11–3.14; verify CPU performance expectations if GPU access is limited.

## Install

pip install nvalchemi-toolkit-ops
uv add nvalchemi-toolkit-ops
poetry add nvalchemi-toolkit-ops

## Installing nvalchemi-toolkit-ops

Before you install: Low friction: pure Python wheel with only numpy and warp-lang as runtime dependencies. Active maintenance (released 10 days ago). Requires Python 3.11–3.14 and CUDA 12 or 13; no system library dependencies beyond what the GPU framework already requires.

License in practice: Apache-2.0 (permissive): you may use, modify, and distribute this package freely in commercial and private projects, provided you retain the license notice and do not hold the authors liable.

Quickstart:

pip install nvalchemi-toolkit-ops

from nvalchemiops.torch.neighbors import neighbor_list

positions = ...  # [num_atoms, 3]
cell = ...  # [num_systems, 3, 3]
pbc = ...  # [num_systems, 3]
edge_index, neighbor_ptr, shifts = neighbor_list(
    positions, cutoff=6.0, cell=cell, pbc=pbc, return_neighbor_list=True
)

Requires CUDA 12 or 13 and Python 3.11–3.14. GPU device recommended for performance; CPU execution support and performance characteristics require verification.

Verify before relying:

- Whether CPU execution is supported and at what performance penalty relative to GPU.
- Exact performance scaling limits and typical throughput on current GPU hardware.
- Whether JAX bindings are feature-complete parity with PyTorch or have limitations.
- Specific CUDA 12 vs. CUDA 13 compatibility and fallback behavior.

## Package facts

- License: Apache-2.0 (permissive)
- Python support: supports_current
- Install friction: low
- Maintenance: active
- Downloads: 238.4K/month (top 15,000 on PyPI)
- Known vulnerabilities: none known

## Tags

gpu molecular dynamics, neighbor list computation, atomistic simulation kernels, batched chemistry operations, warp-lang computational chemistry, differentiable electrostatics, gpu-accelerated dft-d3, gpu-computing, molecular-dynamics, computational-chemistry

[View on SkillFed](https://skillfed.io/packages/nvalchemi-toolkit-ops) · [View on PyPI](https://pypi.org/project/nvalchemi-toolkit-ops/)