cuda-core

What it is and what it does

cuda-core is a Python library that wraps NVIDIA's CUDA core functionality, exposing runtime control, compiler, and linker operations through Pythonic APIs. It bridges Python's ease of use with CUDA's GPU compute performance, allowing developers to manage GPU execution, compilation, and linking directly from Python code without dropping into C/C++.

The package is actively developed by NVIDIA and supports modern Python versions (3.10–3.14) across Linux, Windows, and ARM architectures. It depends on cuda-pathfinder for CUDA toolkit discovery, numpy for numerical operations, and backports.strenum for enum support. Installation requires selecting a CUDA variant (cu12 or cu13) matching your system's CUDA toolkit.

Use it for:

• Manage GPU kernel execution and memory from Python without writing C/C++ code.
• Compile and link CUDA code programmatically as part of a Python build or deployment pipeline.
• Prototype GPU-accelerated algorithms in Python with direct access to CUDA runtime APIs.
• Integrate CUDA operations into data science or machine learning workflows alongside numpy and other Python libraries.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Provides Pythonic APIs to access CUDA core functionalities including runtime control, compiler, and linker operations from Python.

Yes, if you need Pythonic access to CUDA runtime, compiler, or linker operations and have a compatible NVIDIA GPU with CUDA 12 or 13. The package is actively maintained, permissively licensed, has no known vulnerabilities, and is backed by NVIDIA. Install friction is moderate due to platform-specific wheels, but pre-built wheels are available for all major platforms and Python versions.

Install

cuda-core on PyPI

pip install cuda-core

uv add cuda-core

poetry add cuda-core

Installing cuda-core

Before you install

Medium install friction due to platform-specific wheels across multiple Python versions (3.10–3.14) and architectures. Actively maintained with recent commits and no known vulnerabilities.

License in practice

Apache-2.0 permissive license allows commercial and private use with minimal restrictions.

Quickstart

pip install cuda-core[cu12]
import cuda.core
# Access CUDA runtime, compiler, or linker APIs

Requires CUDA 12 or 13 compatible GPU and corresponding CUDA toolkit; Python 3.10 or later.

Verify before relying

• Specific CUDA runtime version requirements beyond the documented cu12/cu13 variants.
• Whether numpy and backports.strenum are runtime dependencies or build-time only.
• Performance overhead of the Python wrapper layer versus direct CUDA C APIs.

Package facts

Evidence: cuda_core-1.1.1-cp310-cp310-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl; cuda_core-1.1.1-cp310-cp310-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl; cuda_core-1.1.1-cp310-cp310-win_amd64.whl; cuda_core-1.1.1-cp311-cp311-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl; cuda_core-1.1.1-cp311-cp311-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl; cuda_core-1.1.1-cp311-cp311-win_amd64.whl; cuda_core-1.1.1-cp312-cp312-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl; cuda_core-1.1.1-cp312-cp312-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl; cuda_core-1.1.1-cp312-cp312-win_amd64.whl; cuda_core-1.1.1-cp313-cp313-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl; cuda_core-1.1.1-cp313-cp313-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl; cuda_core-1.1.1-cp313-cp313-win_amd64.whl; cuda_core-1.1.1-cp314-cp314-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl; cuda_core-1.1.1-cp314-cp314-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl; cuda_core-1.1.1-cp314-cp314t-manylinux_2_24_aarch64.manylinux_2_28_aarch64.whl; cuda_core-1.1.1-cp314-cp314t-manylinux_2_24_x86_64.manylinux_2_28_x86_64.whl; cuda_core-1.1.1-cp314-cp314t-win_amd64.whl; cuda_core-1.1.1-cp314-cp314-win_amd64.whl

Tags

More Libraries packages