Philipp Fischer

> * How did you install pycuda? I installed it simply using `pip install pycuda` ``` C:\WINDOWS\system32>"C:\Program Files (x86)\Python38\python.exe" -m pip install pycuda Collecting pycuda Using cached pycuda-2020.1.tar.gz (1.6 MB)...

It's actually not gpu memory allocation that fails, but the initialization itself. So `import pycuda.autoinit` also fails, because it internally fails in this line: https://github.com/inducer/pycuda/blob/29466d4e93ec20a81ce2534327aed24903c3a2e2/pycuda/autoinit.py#L5 And all this seems to...

`_driver.cp38-win32.pyd` seems to crash before it even get's to loading cuda and calling `cuInit()`. I will have to investigate deeper. The following call is out of address space, I just...

Honestly, I was a bit to lazy to get the whole build environment ready to compile this myself, so I downloaded and installed this pre-compiled wheel `pycuda-2020.1+cuda102-cp38-cp38-win32.whl` from https://www.lfd.uci.edu/~gohlke/pythonlibs/#pycuda and...

Hmm interesting. So this means that during `pip install pycuda` it must have been compiled locally on my machine and something must have gone wrong with it?

Thanks, I understand. I think I would need to create a new PyPI package from the `utils.py` file that does the SIP things. But let me first check if some...

> 1. The way of connecting to a device should be encapsulated in a separate lib. This way HA only contains code for calling the device, not _how_ to call...

Adding link to docs PR, hoping the pipeline will pick up the link: https://github.com/home-assistant/home-assistant.io/pull/30396 to get rid of the `docs-missing` label.

@joostlek , can we proceed?

Can anyone else jump in? I think Joostlek is busy. @zweckj @autinerd