NVIDIA Warp is an open-source developer framework designed for building and accelerating data generation and spatial computing in Python. It allows developers to write GPU-accelerated, kernel-based programs for simulation, AI, robotics, and machine learning (ML). With Warp, Python developers can create GPU-accelerated, 3D simulation workflows that drive ML pipelines in PyTorch, JAX, PhysicsNeMo, and NVIDIA Omniverse. Warp offers simulation performance equivalent to native CUDA code, combined with the convenience and productivity of Python.

Features

• Kernel-Based Code: Performs just-in-time (JIT) runtime compilation of Python functions to x86 and CUDA kernel-level code, providing fine-grained control over threads and native support for conditional logic.
• Differentiable Programming: Generates reverse-mode (adjoint) kernels that propagate gradients of simulation results back into frameworks like PyTorch and JAX for network training, design optimization, and parameter estimation.
• Built for Spatial Computing: Includes built-in functionality for 3D simulation and geometry processing, with higher-level data structures such as meshes, hash grids, and sparse volumes (NanoVDB) for GPU-accelerated geometric queries.
• Integration with NVIDIA Technologies: Supports PyTorch, JAX, PhysicsNeMo, and NVIDIA Omniverse, enabling seamless integration into existing workflows.

Benefits

• Efficiency: Achieves simulation performance equivalent to native CUDA code, significantly speeding up development.
• Flexibility: Provides a low-level abstraction that maps closely to GPU hardware, offering fine-grained control over threads and native support for conditional logic.
• Productivity: Combines the convenience and developer productivity of Python with the performance of CUDA.
• Scalability: Supports scalable texture compression for various GPU types, making it suitable for large-scale simulations.
• Integration: Easily integrates with existing ML and simulation frameworks, enhancing workflow efficiency.