cuda-bindings
Python bindings for CUDA
Decision gist · record as of 2026-08-14
Yes, if you need direct CUDA API access from Python and accept the custom NVIDIA license. The package is actively maintained with no known vulnerabilities and widely used. Medium install friction is manageable for GPU-focused projects. Verify the license terms for your use case first.AI-flagged interpretation of the facts on this page — verify before relying
Before you install
- Requires CUDA Toolkit to be installed on the system; cuda-pathfinder runtime dependency must be available.
- Medium install friction due to platform-specific wheels across multiple architectures and Python versions.
- Active maintenance with recent releases and 3341 repository stars indicate ongoing support.
License · maintenance · safety
LicenseRef-NVIDIA-SOFTWARE-LICENSE (unclear) — License treatment is unclear—the package uses a custom NVIDIA SOFTWARE LICENSE (LicenseRef-NVIDIA-SOFTWARE-LICENSE) rather than a standard open-source license. Review the license terms before use in proprietary or redistributed projects.
last release 2026-05-29 (77 days) · last repo commit 2026-08-14 · 3,341 stars
0 known vulnerabilities (OSV.dev, 2026-08-14) · 50,138,807 downloads/mo, #575 on PyPI
Alternatives
Verify before relying
pip install cuda-bindings
import cuda
# Access CUDA host APIs directly
# Example: cuda.cudaGetDeviceCount()- Whether the custom NVIDIA SOFTWARE LICENSE permits commercial use or redistribution.
- Specific CUDA Toolkit version requirements beyond what cuda-pathfinder provides.
- Performance characteristics compared to direct C API calls.
What it is and what it does
cuda-bindings is a Python library that exposes CUDA host APIs with 1:1 coverage, allowing developers to call CUDA functions directly from Python without writing C extensions. It is maintained by NVIDIA and provides full access to the CUDA runtime and driver interfaces, making it suitable for applications that need low-level GPU control or integration with existing CUDA workflows.
The package is distributed as platform-specific wheels for Python 3.10 through 3.14 on x86_64, aarch64, and Windows architectures. It depends on cuda-pathfinder at runtime to locate the CUDA installation. The library is actively maintained, with recent releases and comprehensive documentation, though its custom license requires careful review before use in commercial or redistributed contexts.
Use it for
- Direct GPU device management and querying (device count, properties, selection) from Python.
- Low-level CUDA memory operations (allocation, transfer, deallocation) without C extension overhead.
- Integration of CUDA workflows into Python applications that require fine-grained GPU control.
- Building higher-level GPU libraries or frameworks that need direct access to CUDA runtime APIs.
- Prototyping and testing CUDA functionality in Python before porting to production C/C++ code.
Worth the install?
AI-flagged interpretation of the facts on this page. Verify before relying on it.
Yes, if you need direct CUDA API access from Python and accept the custom NVIDIA license.
The package is actively maintained with no known vulnerabilities and widely used. Medium install friction is manageable for GPU-focused projects. Verify the license terms for your use case first.
Install
cuda-bindings on PyPI
Before you install
Medium install friction due to platform-specific wheels across multiple architectures and Python versions. Active maintenance with recent releases and 3341 repository stars indicate ongoing support.
Requires CUDA Toolkit to be installed on the system; cuda-pathfinder runtime dependency must be available.
License in practice
License treatment is unclear—the package uses a custom NVIDIA SOFTWARE LICENSE (LicenseRef-NVIDIA-SOFTWARE-LICENSE) rather than a standard open-source license. Review the license terms before use in proprietary or redistributed projects.
Quickstart
pip install cuda-bindings
import cuda
# Access CUDA host APIs directly
# Example: cuda.cudaGetDeviceCount()Verify before relying
- Whether the custom NVIDIA SOFTWARE LICENSE permits commercial use or redistribution.
- Specific CUDA Toolkit version requirements beyond what cuda-pathfinder provides.
- Performance characteristics compared to direct C API calls.
Package facts
| License | LicenseRef-NVIDIA-SOFTWARE-LICENSE unclear |
| Python support | Supports the current Python release >=3.10 |
| Install friction | Medium. Platform-specific wheel |
| Runtime dependencies | 1 packagecuda-pathfinder |
| Maintenance | Actively maintained 77 days since the last release |
| Last repo commit | |
| First released | |
| Downloads | 50,138,807 / month, #575 on PyPI 30-day window, as of 2026-08-14 |
| Known vulnerabilities | None known OSV.dev, checked 2026-08-14 |
| Classifiers | Environment :: GPU :: NVIDIA CUDAIntended Audience :: DevelopersProgramming Language :: PythonProgramming Language :: Python :: 3.10Programming Language :: Python :: 3.11Programming Language :: Python :: 3.12Programming Language :: Python :: 3.13Programming Language :: Python :: 3.14Topic :: DatabaseTopic :: Scientific/Engineering |
Evidence: cuda_bindings-13.3.1-cp310-cp310-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl; cuda_bindings-13.3.1-cp310-cp310-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl; cuda_bindings-13.3.1-cp310-cp310-win_amd64.whl; cuda_bindings-13.3.1-cp311-cp311-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl; cuda_bindings-13.3.1-cp311-cp311-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl; cuda_bindings-13.3.1-cp311-cp311-win_amd64.whl; cuda_bindings-13.3.1-cp312-cp312-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl; cuda_bindings-13.3.1-cp312-cp312-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl; cuda_bindings-13.3.1-cp312-cp312-win_amd64.whl; cuda_bindings-13.3.1-cp313-cp313-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl; cuda_bindings-13.3.1-cp313-cp313-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl; cuda_bindings-13.3.1-cp313-cp313-win_amd64.whl; cuda_bindings-13.3.1-cp314-cp314-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl; cuda_bindings-13.3.1-cp314-cp314-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl; cuda_bindings-13.3.1-cp314-cp314t-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl; cuda_bindings-13.3.1-cp314-cp314t-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl; cuda_bindings-13.3.1-cp314-cp314t-win_amd64.whl; cuda_bindings-13.3.1-cp314-cp314-win_amd64.whl
Tags
Let your AI agent find packages like this
Example. Real query, live index.
You found this page by searching. An agent finds it by wishing: SkillFed indexes 14,416 PyPI packages by what they can do, searchable in plain language.
wish › “low-level cuda interface”
- cuda-bindingsProvides low-level Python bindings for CUDA host APIs, offering 1:1…
- cuda-pythoncuda-python is a metapackage providing Pythonic access to NVIDIA's…
- cupti-pythonProvides Python bindings to NVIDIA's CUPTI profiling and tracing APIs…
Give your agent the search over MCP, or paste the wish link into any chat.
More Scientific/Engineering packages
NumPy provides an N-dimensional array object and a comprehensive suite of mathematical, linear algebra, Fourier transform, and random number functions for scientific computing in Python.
pandas provides fast, flexible data structures (Series and DataFrame) for loading, cleaning, transforming, and analyzing labeled or relational data in Python.
scipy provides numerical algorithms for mathematics, science, and engineering—including optimization, integration, linear algebra, Fourier transforms, signal and image processing, and ODE solvers—built on numpy arrays.
scikit-learn provides a comprehensive Python library for supervised and unsupervised machine learning, including classification, regression, clustering, dimensionality reduction, and model evaluation tools built on NumPy and SciPy.
Install it if you need to train, evaluate, or deploy supervised or unsupervised learning models.
dill extends Python's pickle module to serialize and deserialize a much wider range of Python objects, including functions, lambdas, classes, and interpreter sessions, to byte streams for storage or network transmission.
Multiprocess is an enhanced fork of Python's standard multiprocessing library that uses dill for better serialization, allowing you to spawn processes with a threading-like API and share complex objects between them.
Install it if you use multiprocessing and encounter pickle serialization limits with lambdas or complex objects.
See also cuda-pathfinder · cuda-python · cuda-core · nv-ingest-client · hip-python · cuda-tile · pycuda · nvidia-cuda-crt · nvidia-cuda-runtime · nvidia-nvfatbin