cudadrv.py loads nvcuda.dll and nvml.py loads nvml.dll on win32, but everything so far has been validated on linux (and the test suite fakes the libraries, so ci can't catch a loading problem).
if you're on windows with an nvidia card, running:
pip install kernelmeter
kernelmeter info
and reporting whether the attribute table and the (nvml) lines show up would be enough. nvml.dll sometimes lives in System32 rather than on PATH, so the fallback list may need a program-files path added.
cudadrv.py loads nvcuda.dll and nvml.py loads nvml.dll on win32, but everything so far has been validated on linux (and the test suite fakes the libraries, so ci can't catch a loading problem).
if you're on windows with an nvidia card, running:
and reporting whether the attribute table and the (nvml) lines show up would be enough. nvml.dll sometimes lives in System32 rather than on PATH, so the fallback list may need a program-files path added.