Is it possible to get direct access to Nvidia (for running TensorFlow, etc.) in a Linux virtual machine from under Windows?

Question

virtual box

Good afternoon!
The main environment in Windows. A little (hobby) I'm starting to look towards machine learning.
Previously, in virtual machines, I never forwarded equipment from the host OS to the virtual one.
Will Nvidia GeForce 1060 forwarding work on Linux installed on a virtual machine (hereinafter CUDA + Tensorflow + related ones)? What problems can there be?
Which virtualization OS is better to serve, who installed it - what problems would there be?
Please do not offer to install everything on the host Windows10, I would like it to be on a virtual machine with Linux.
Thanks in advance.

Answer 1

No
A full-fledged working forwarding does not work, even vice versa - linux is the host, Windows is the guest. There are a lot of success stories in tyrnet about how it was done, but you can’t repeat it. Not least because nvidia is opposed to such things - they believe that for this you need to buy non-epically expensive K1 / K2.
I have never heard of forwarding from Windows anywhere, IMHO this is impossible in principle. Forwarding from linux seems to work if the vidyuha is from ATI