I haven’t looked into how to configure this but it should be possible, and you would use the motherboard HDMI port for the VM, and the ports on the dGPU for the host. As usual, the arch wiki is your friend, even if you are not using arch

But… If you don’t care about VM performance (seeing as you are passing the iGPU to it) you should look into other options like virtio or sr-iov, so you don’t need to fiddle with the HDMI ports. Please notice that virtio is paravirtualized and only works well for Linux guests, and sr-iov is real hardware virtualization and requires hardware support. Both these methods require only one GPU. Once again, look at the arch wiki and the qemu wiki.

Also, if you are using Linux guests, you should really look into “GPU native context” which is a paravirtualization method that works similarly to Hyper-V’s GPU paravirtualization (which is currently the best) and would allow almost native performance for the VM, without requiring multiple GPUs. It is not available for amdgpu yet, but you can follow development here.

P.s. if you are using windows hosts, paravirtualization methods will not be satisfactory for the foreseeable future. You will need either passthrough (like you suggested) or full virtualization (with sr-iov). I can give you more details if you like.

I have the same processor and an RX570. I can pass through the dGPU to the VMs and keep the iGPU for the host, but the reverse was not possible for me.

Does the motherboard matter for this? Mine is not so fancy.

deleted by creator

It depends on what else is part of that iommu group. Because you can’t just pass through single PCI device you have to pass through everything that is part of that group