Warp-OptiX: Ray Tracing with NVIDIA Warp

Documentation | Examples

Warp-OptiX brings NVIDIA OptiX ray tracing to Warp. Write ray-generation, hit, miss, and callable programs as Warp kernels, or use the included path tracer. The Python API provides managed contexts, acceleration structures, automatic shader binding tables, and launches that mirror wp.launch.

Installation

pip install warp-optix

GPU execution requires a compatible NVIDIA GPU and driver; no CUDA Toolkit or OptiX SDK is needed. Wheels support Python 3.12 on Linux x86-64 and Windows x64. Source builds require Python 3.12 and a C++17 compiler. See the installation guide for source and development setup.

PyOptiX 9.1.1 is installed automatically. The package includes OptiX SDK 9.1.0 headers for runtime kernel compilation, so a separate OptiX SDK installation is not required.

Optional extras add path tracing, UI, USD loading, and neural-network training:

pip install "warp-optix[pathtracing,ui,usd]"
pip install "warp-optix[training]"

Getting started

import warp as wp
import warp_optix as wo
from warp_optix.pathtracing import PathTracer

Follow the API guide to build a custom ray tracer, or explore the runnable examples. Advanced applications can access PyOptiX and supplemental bindings through warp_optix.native.

CUDA graph capture is currently unsupported for OptiX operations. Submit OptiX work between captured Warp graphs on the same stream; CUDA stream ordering is supported.

Support and contributions

Report bugs and feature requests through GitHub Issues. The project is currently not accepting outside contributions; see the contribution policy for updates. Report security vulnerabilities privately using the security policy.

License

Warp integration uses Apache-2.0; supplemental bindings use BSD-3-Clause. Bundled OptiX headers retain their NVIDIA SDK terms. See third-party notices for component licenses and source provenance. License texts accompany the distribution.