Running Numba on AMD GPUs using PyOpenCL

Todo

This article is a work-in-progress

Preamble

It is unfortunate that NVIDIA rules the GPU world when it comes to data engineering. Especially where Python is concerned. For what it’s worth, the Anaconda Foundation has been backing projects such as the excellent Numba library that provides an LLVM compiler for Python, enabling Just-In-Time compiling for Python code.

Numba works natively for NVIDIA GPUs, but NVIDIA’s support for Linux is apalling to say the least.

I believe that it is feasible to setup numba to work with PyOpenCL instead of PyCUDA, and this will allow developers to use AMD GPUs instead.

---