What is the relationship between this and pycuda ? about cuda-python

[leofang]

It's worth adding that PyCUDA is the ancestor of Python GPU programming. In particular, it showed the combination of

• exposing CUDA C APIs and programming interface to Python
• providing a GPU array container

brings the full capability of CUDA to Python users. In a sense Numba, CuPy, and now CUDA Python are successors of it, specialized in each subdomain. I don't think it's exaggerating to say without PyCUDA's success it'd be slower to see the booming in growth of Python GPU users.

But, CUDA Python provides almost full coverage of the driver/runtime/NVRTC APIs, which is something none of the projects had the capacity to offer due to the wide scope. Also, CUDA Python is an official NVIDIA product 🙂

[shwina]

Hi @jinluyang

First, PyCUDA is very much open source: https://github.com/inducer/pycuda

PyCUDA is a larger library, offering a Pythonic interface to the driver API, as well as higher level features such as GPU arrays and utilities for metaprogramming.

CUDA Python on the other hand exclusively provides Python bindings for the CUDA driver and runtime APIs.

So yes - there is definitely some overlap between the two. However CUDA Python is much more narrow and focused in its scope.