Nvidia has officially made its Warp Python framework open source under the Apache 2.0 license. Designed for high-performance physical simulations, data generation, and spatial computing, Warp compiles Python functions just-in-time into kernel code that can run on both x86 CPUs and CUDA-enabled GPUs.

Although Nvidia hasn’t publicly explained the reasons behind the license switch, the move follows criticism from the developer community regarding the framework’s original license. Previously, Warp was released under a proprietary license that, while disclosing the source code, retained significant usage restrictions. The transition to Apache 2.0 now qualifies Warp as fully open source, addressing those concerns and encouraging broader adoption.

Performance Close to Native CUDA Code

Warp offers a specialized set of tools—or primitives—for use in physics-based simulations, robotics, sensor applications, and geometric processing. These include advanced data structures such as meshes, hash grids, and sparse volumes optimized for GPU use. Unlike tensor-based programming models, Warp provides developers with more direct control over kernel and thread management, built-in conditional logic, and support for sparse scatter/gather operations, enhancing vectorized computing performance.

The framework supports reverse-mode differentiation, enabling seamless integration into machine learning environments like PyTorch, JAX, Nvidia Omniverse, PhysicsNeMo, and Paddle. Nvidia promotes Warp as delivering simulation speeds comparable to native CUDA code, while maintaining the ease and developer productivity of Python.

Platform and GPU Support

Warp is compatible with x86-64 and ARMv8 architectures and runs on Windows, Linux, and macOS. To leverage GPU acceleration, users need CUDA-capable GPUs with compatible drivers, starting from the GeForce GTX 900 series. For Python, Nvidia recommends version 3.9 or later. The binaries available via PyPI are built with CUDA 12 and require driver versions 525.60.13 or newer for Linux (x86-64), and 528.33 or newer for Windows (x86-64).

By going open source, Nvidia positions Warp as a more accessible and powerful tool for developers working on GPU-accelerated simulations and machine learning applications, solidifying its role in the broader scientific and AI programming ecosystem.