Yuri, who has written a script and tutorial for single GPU passthrough, writes:

Single GPU passthrough and the way a graphics card is attached to Xorg is still a little experimental.After booting the VM, the Linux host will no longer have a graphics device, so one cannot switch back and forth between the two.The graphics card must be reset before it can be used by another windows manager/OS – some cards behave better than others.As simple as it sounds, this approach has it’s downsides: When the Windows VM is shutdown, a program or script resets the graphics card and re-attaches it to the Linux host. Single GPU passthrough uses one graphics card for the Linux host, which is passed through to the Windows VM when needed. The most common approach is “single GPU passthrough”, and several tutorials have focused on that subject. One approach is to make the (powerful) discrete GPU available to Linux/Xorg. The IGD or a low-power GPU will handle the less demanding jobs of web surfing, video/movie display and a whole bunch of productivity tasks just fine.īut what if you wish to play a game under Linux? Numerous Linux-based games available today already have higher demands on the graphics device. Typically you assign the IGD or internal graphics device (inside the CPU) to the host and the (powerful) discrete graphics card to the Windows VM. Using KVM and GPU passthrough – also termed “VFIO” – allows users to play graphic intensive games inside a virtual machine (VM) by using a dedicated GPU for the Windows VM. Given the right configuration, this “Windows virtual gaming machine” will have little to no performance penalty compared to a native (bare metal) Windows installation. Here and elsewhere Linux users run their Microsoft Windows OS as a virtual machine on Linux and pass through a dedicated GPU.

Windows vm on linux