NVIDIA/warp

Health v2 · maintenance only

Adoption not part of the score

7033 stars · 602 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

NVIDIA Warp is a Python framework that JIT-compiles regular Python functions into efficient CPU or CUDA GPU kernels. It provides differentiable primitives for physics simulation, robotics, geometry processing, and machine-learning pipelines with PyTorch, JAX, and Paddle.

Use cases

• simulate millions of particles on the GPU from Python
• write differentiable physics simulations for robotics
• accelerate geometry processing kernels with CUDA
• integrate custom GPU kernels into PyTorch or JAX training pipelines
• prototype tile-based GPU programs in Python
• run differentiable optimization like shape optimization or inverse problems

When to choose

• you need GPU-accelerated, differentiable simulation code written in Python
• you want custom CUDA-like kernels without writing C++/CUDA
• you're building robotics or physics ML pipelines needing gradients through simulation

When to avoid

• you only need general tensor math without simulation primitives (use NumPy or PyTorch)
• you need GPU acceleration on macOS (Warp supports CPU only there)
• your project requires non-NVIDIA GPU acceleration

Facets

framework · maturity active

machine-learning simulation gpu-computing compiler math machine-learning simulation robotics gpu-computing graphics python windows differentiable-programming jit-compilation physics-simulation kernels pytorch-integration jax-integration nvidia linux macos gpu cuda

Member repositories

For agents

markdown · JSON · MCP: product_card(name="NVIDIA/warp")