This is a nice work method. But my biggest grip with it is that I use the GPU for CUDA calculations. At least in OSX it's not possible to directly access the G

This is a nice work method. But my biggest grip with it is that I use the GPU for CUDA calculations. At least in OSX it's not possible to directly access the GPU trough a VM (in Vagrant or any other way), while in Linux it's possible with KVM. How does it work on Windows?