articleschannelstagsspacestoolkit
NVIDIA Technical Blog

Just Released: NVIDIA Warp is Now Open-Source Under Apache 2.0

View Original
thumbnail

NVIDIA Warp Open-Source Release Summary

  • Introduction: NVIDIA Warp is now open-source under Apache 2.0 license. It is a Python framework for high-performance simulation and graphics code.
  • Binaries: Currently built with CUDA 12 runtime, requiring a minimum CUDA driver version of 525.60.13 (Linux x86-64) or 528.33 (Windows x86-64).
  • Nightly Builds: Available on the NVIDIA Package Index, can be installed using provided commands. Not published for macOS.
  • CUDA Requirements: Packages built with CUDA Toolkit 11.x require NVIDIA driver 470 or newer, while those built with CUDA Toolkit 12.x need driver 525 or newer.
  • Dependencies: Users need to ensure installation of certain packages before running examples.
  • Examples: Can be run from the command-line, some require CUDA-capable devices and are not recommended for Preview in macOS.
  • Building from Source: Provides instructions for building Warp from source using compatible CUDA Toolkit and installing the package.

Nightly Builds Installation Command

pip install --index-url https://pypi-nightly.ngc.nvidia.com nvidia-warp

Setting NVIDIA's Package Repository as an Extra Index

Add the following line to your environment to simplify future installations:

export PIP_EXTRA_INDEX_URL=https://pypi-nightly.ngc.nvidia.com

Building Warp from Source Instructions

  • Update the driver to meet CUDA requirements.
  • Build Warp from source using a compatible CUDA Toolkit.
  • Install the built Warp package using provided instructions.

Running Examples

  • Ensure installation of required packages.
  • Run examples from the command-line and browse source code.
  • Note device requirements for certain examples and limitations in macOS Preview.