With the Windows VM are you doing GPU pass through to get native performance? Is there still a relatively minimal overhead doing it that way? I would be interested in running applications in their own Windows VM(one at a time at least) but the VM is essentially invisible and only application window is available?