As someone who runs an R9 Fury PCIe passthrough setup, I doubt that setup was really practical or stable. Fiji has PCI reset issues if you try something like this, you'd have to shutdown every VM and reboot the host in order to restart just one VM. Driver updates will also hard freeze the host if you don't use a workaround.
Well, there's a way around this, but it requires a virtual UEFI project originally for Fedora Rawhide and you'd need to flash each and every GPU with a modded bios that supports GOP, because IIRC the Nano doesn't out of the box.
There's a huge list of other considerations too.
Then again, I'm not familiar with the proprietary software Linus is using. It's possible that it sorts out some of these issues (but I doubt it).
If you give the VM the vbios of the card, you can reboot the VM without rebooting the host.
This only works as a qemu command line argument.-device vfio-pci,host=07:00.0,......,romfile=/path/to/your/gpu/bios.bin
I saw that suggested, didn't work with a Fury with SeaBIOS for me. I think it's a Fiji issue, as I've heard your suggested fix worked for Hawaii?
The only thing that fixed the reset issue for me was using OVMF.