(Creating a new post so as to catch wider attention) As per RFC: Moving the CUDA target to a new package maintained by NVIDIA, numba-cuda
packages are now available - please try them out in your environment and report any issues on the numba-cuda
issue tracker or post back here for further discussion.
Install with conda:
conda install nvidia::numba-cuda
Install with pip:
pip install numba-cuda
If all is well, there should be no perceptible difference between using the out-of-tree target (numba-cuda
) and the built-in (numba
) one, except you can observe where the cuda
package is loaded from. By executing
python -c "from numba import cuda; print(cuda.__file__)"
you would see
<site-packages>/numba/cuda/__init__.py
with the built-in CUDA target, or
<site-packages>/numba_cuda/numba/cuda/__init__.py
with the numba-cuda
target.
A conda-forge recipe / package is still forthcoming.