A Python framework for GPU-accelerated simulation, robotics, and machine learning.

GitHub commit activity GitHub - CI

NVIDIA Warp

Warp is a Python framework for GPU-accelerated simulation, robotics, and machine learning. Warp takes regular Python functions and JIT compiles them to efficient kernel code that can run on the CPU or GPU.

Warp comes with a rich set of primitives for physics simulation, robotics, geometry processing, and more. Warp kernels are differentiable and can be used as part of machine-learning pipelines with frameworks such as PyTorch, JAX and Paddle.

 ![](https://github.com/NVIDIA/warp/raw/main/docs/img/header.jpg)

A selection of physical simulations computed with Warp

Quick Start

Simulate one million particles under gravitational attraction, in 20 lines:

import warp as wp
import numpy as np

num_particles = 1_000_000
dt = 0.01


@wp.kernel
def gravity_step(pos: wp.array[wp.vec3], vel: wp.array[wp.vec3]):
    i = wp.tid()
    position = pos[i]
    dist_sq = wp.length_sq(position) + 0.01  # softened distance
    acc = -1000.0 / dist_sq * wp.normalize(position)  # gravitational pull toward origin
    vel[i] = vel[i] + acc * dt
    pos[i] = pos[i] + vel[i] * dt


rng = np.random.default_rng(42)
positions = wp.array(rng.normal(size=(num_particles, 3)), dtype=wp.vec3)
velocities = wp.array(rng.normal(size=(num_particles, 3)), dtype=wp.vec3)

for _ in range(100):
    wp.launch(gravity_step, dim=num_particles, inputs=[positions, velocities])

print(positions.numpy())

Installing

Warp requires Python 3.10 or newer. We publish warp-lang wheels on PyPI for Windows (x86-64), Linux (x86-64 and AArch64), and macOS (Apple Silicon). The Windows x86-64 and Linux wheels support CPU execution and CUDA acceleration. The macOS wheels support CPU execution but not Metal acceleration.

PyPI and nightly wheels for Linux and Windows use CUDA Toolkit 13.4. They require an NVIDIA R580-series or newer driver and a Turing (sm_75) or newer GPU for CUDA acceleration. For CUDA 12 environments, download a +cu12 wheel from GitHub Releases or build Warp from source with CUDA 12.

Windows on Arm supports experimental CUDA source builds with CUDA Toolkit 13.4 or newer; see the Windows on Arm build instructions.

The easiest way to install Warp is from PyPI:

pip install warp-lang

You can also use pip install warp-lang[examples] to install additional dependencies for running examples and USD-related features.

For nightly builds, conda, CUDA 12 compatibility wheels, building from source, and CUDA driver requirements, see the Installation Guide.

Tutorial Notebooks

The NVIDIA Accelerated Computing Hub also hosts Warp tutorial notebooks that can be opened in Colab:

Notebook	Colab Link
01. Introduction to NVIDIA Warp
02. Ising Model
03. 2D Navier-Stokes Solver
04. Differentiable Simulations in Warp

Running Examples

The warp/examples directory contains examples covering physics simulation, geometry processing, optimization, and tile-based GPU programming. Before running examples, install the optional example dependencies using:

pip install warp-lang[examples]

On Linux aarch64 systems (e.g., NVIDIA DGX Spark), the [examples] extra automatically installs usd-exchange instead of usd-core as a drop-in replacement, since usd-core wheels are not available for that platform.

Examples can be run from the command-line as follows:

python -m warp.examples..

Most examples can be run on either the CPU or a CUDA-capable device, but a handful require a CUDA-capable device. These are marked at the top of the example script. Some examples generate USD files containing time-sampled animations in the current working directory. These can be viewed in Pixar's UsdView, Blender, or any USD-compatible viewer.

To browse the example source code, you can open the directory where the files are located like this:

python -m warp.examples.browse

warp/examples/core

dem

fluid

graph capture

mesh

nvdb

raycast

raymarch

sample mesh

sph

torch

wave

2-D incompressible turbulence in a periodic box

3-D FDTD Luneburg lens

warp/examples/geometry

isosurface

sparse marching cubes

warp/examples/fem

diffusion 3d

mixed elasticity

apic fluid

streamlines

distortion energy

taylor green

kelvin helmholtz

magnetostatics

adaptive grid

nonconforming contact

darcy level-set optimization

elastic shape optimization

cantilever topology optimization

warp/examples/optim

diffray

fluid checkpoint
basic · custom backward

particle repulsion

navier-stokes perturbation

warp/examples/tile

mlp

nbody

mcgp

Learn More

Please see the following resources for additional background on Warp:

Support

See the FAQ for common questions.

Problems, questions, and feature requests can be opened on GitHub Issues.

For inquiries not suited for GitHub Issues, please email .

Contributing

Contributions and pull requests from the community are welcome. Please see the Contribution Guide for more information on contributing to the development of Warp.

License

Warp is provided under the Apache License, Version 2.0. Please see LICENSE.md for full license text.

This project will download and install additional third-party open source software projects. Review the license terms of these open source projects before use.

Building from Source

When building Warp from source using the build_lib.py script, the build process automatically downloads NVIDIA libmathdx. Pre-built Warp packages (e.g., from PyPI) already include libmathdx statically linked into the library binaries. In both cases, libmathdx is governed by the NVIDIA Software License Agreement.

NOTICE AND DISCLAIMER: This software automatically retrieves, accesses or interacts with external materials. Those retrieved materials are not distributed with this software and are governed solely by separate terms, conditions and licenses. You are solely responsible for finding, reviewing and complying with all applicable terms, conditions, and licenses, and for verifying the security, integrity and suitability of any retrieved materials for your specific use case. This software is provided "AS IS", without warranty of any kind. The author makes no representations or warranties regarding any retrieved materials, and assumes no liability for any losses, damages, liabilities or legal consequences from your use or inability to use this software or any retrieved materials. Use this software and the retrieved materials at your own risk.

Publications & Citation

Research Using Warp

Our PUBLICATIONS.md file lists academic and research publications that leverage the capabilities of Warp. We encourage you to add your own published work using Warp to this list.

Citing Warp

If you use Warp in your research, please use the "Cite this repository" button on the GitHub repository page or refer to the CITATION.cff file for citation information.