I occasionally need to work on scientific software that uses CUDA and is written in Python.
If possible, I would like to do that on an iMac 2019 (Intel, with a Radeon Pro 580X, running macOS Monterey and Anconda as a Python package manager). Of course, that GPU is not able to support CUDA drivers.
But is there a way to install a CUDA toolkit and have it use a CPU backend, just to be able to build and analyse those applications? I don‘t care about performance at all for my use case. But NVidia seems to have dropped macOS support for their more current CUDA toolkits, and packages such as
tensorflow-gpu don‘t even exist for macOS as a platform on the Anaconda repositories…