VGA passthrough and desktop virtualization

Question

In short, my dream is to have one machine with multiple paravirtualized desktop and server guests, one of which has to be a Windows desktop with powerful graphics.

As Windows can't be paravirtualized in the normal sense of the word, I was quite happy when I heard about VGA passthrough, but then I read on. As I understand it, such a setup would mean that the Graphics will be dedicated to one particular guest, thus you wouldn't be able to switch between guests.

If this is in fact so, would someone please explain to me what the purpose/use of VGA passthrough is? I can think of no real use for it. Yes it's a cool technology, but to me it seems pointless.

It's true that it's possible to passthrough individual VMs to separate GPUs, which is also cool, but in the end I should think that two seperate computers would make life a whole lot simpler. Again it seems rather pointless.

Answer 1

Device passthrough is based on IOMMU by re-mapping device addresses and providing to the guest. Hardware can be mapped by just one machine, physical or virtual, so if your guests needs direct I/O to some hardware, you should have it 1:1. IOMMU must be supported by CPU (search for Intel VT-d or AMD-Vi), chipset, BIOS and OS + VM solution.

Originally IOMMU was primarly for passing NICs, HDD controllers, etc. With VGA pass-through you can meet some problems, it's little more complicated, but possible. Do some research, you can start here. You can only solve this with more than one VGA, because that one is already initialized.