RAPIDS interop with PyCuda?

Hi, first post here–so let me know if there was a better place to post this. I’m excited about the capabilities that the RAPIDS team is bringing to Python. I used some CUML capabilities, even made my own kernels on the data–works great. But after some initial development, I found myself needing to go deeper into the CUDA library–specifically I wanted to call out to cuSolver.

Traditionally it looks like PyCuda is a good way for Python users to reach cuSolver… but this brings me to the fundamental question (which isn’t clear to me). Can RAPIDS interoperate with PyCuda? Can I pass a cuDF data frame from RAPIDS for use in PyCuda without churning in and out of device memory? Any guidance appreciated.

Please checkout CuPy. There’s a good chance what you need it built-in.

  1. https://github.com/cupy/cupy
  2. https://docs.cupy.dev/en/stable/