That seems to imply that it can use the same GPU as the host? I thought a dedicated one was required (not just a 'common use').
I haven't worked in windows space in awhile though, and state changes rapidly, so there may be other options now that utilize a single GPU across host and vm without losing performance.