1) Just the eGPU, PCIe is not passed through
2) It depends, you'll have to test
3) Yes, because Apple only supports eGPU
4) Not a fusion question, but if it has enough wattage, yes.
For the second question, we tried the usb device can be used for both host and guest.
and if you connect a network by usb adapter, it works well.
My question in regards to 3) can more clearly be reformulated as: Will i be able to use a NVIDA GPU in an eGPU enclosure and pass the GPU through to the guest OS?
MacOS only allow for AMD GPU in eGPU enclosures but OSs like Windows or Linux work fine with NVIDIA.
No. Because Apple only supports AMD GPU's, and doesn't make any others available to the host OS to be virtualized (it's not passed through).
As I just installed Fusion 12 to test egpu specifically, the VM doesn't recognize the actual GPU, Vega FE, so what driver is it using for the egpu?
Or does it just pass the graphics through and use the OS X driver?
It appears they need to provide much better documentation on this functionality as it could really set them apart from Parallels, Vbox, etc...
It uses VMware Tools drivers, and the SVGA II virtual graphics device.
What's cool is because it's abstracted, you can use it with or without the eGPU.
You can plug in while Fusion is running, and it will 'just work' and start using the eGPU. I demo it in my upcoming VMworld presentation, HCP1833 btw
The VM is still compatible with the rest of the VMware infrastructure, so you can move it to workstation or ESXi and not have to 'disconnect' the eGPU in the guest like you would if Passthrough were possible.
So as far as docs go it's like... 'plug in, it works'. Not a great idea to disconnect it while running (I have had it both be fine and also crash), but you can just suspend, disconnect, resume and keep going.