If you're referring to GPU passthrough, then yes, in practice much better with AMD cards then Nvidia cards.
Nvidia no longer throws error 43 on new drivers. AMD cards before the 6000 series also have a PCIe reset bug which forces you to reboot the host.
I have been wondering... Instead of exposing the hardware GPU to the appVM, would a Vulkan virtualization work? I.e., VM sees a Vulkan API that forwards calls to a dedicated graphics VM that runs them.
(I understand that getting windos to use them could be hard.)
I don't have any direct knowledge but those kind of solutions create security problems.