Installation

Runtime Requirements

cuda.core is supported on all platforms that CUDA is supported. Specific dependencies are as follows:

	CUDA 11	CUDA 12
CUDA Toolkit ¹	11.2 - 11.8	12.0 - 12.6
Driver	450.80.02+ (Linux), 452.39+ (Windows)	525.60.13+ (Linux), 527.41+ (Windows)

cuda.core supports Python 3.9 - 3.13, on Linux (x86-64, arm64) and Windows (x86-64).

cuda.core works with cuda.bindings (part of cuda-python) 11 or 12. For example with CUDA 12:

$ pip install cuda-core[cu12]

and likewise use [cu11] for CUDA 11.

Note that using cuda.core with NVRTC or nvJitLink installed from PyPI via pip install requires cuda.bindings 12.8.0+ or 11.8.6+.

Same as above, cuda.core can be installed in a CUDA 11 or 12 environment. For example with CUDA 12:

$ conda install -c conda-forge cuda-core cuda-version=12

and likewise use cuda-version=11 for CUDA 11.

Note that to use cuda.core with nvJitLink installed from conda-forge currently requires it to be separately installed:

$ conda install -c conda-forge libnvjitlink

(can be combined with the command above). This extra step will be removed in a future release.

$ git clone https://github.com/NVIDIA/cuda-python
$ cd cuda-python/cuda_core
$ pip install .

For now cuda-python (cuda-bindings later) 11.x or 12.x is a required dependency.