Then you will probably use some operations layer on top of that (e.g. QubesOS and XCP-NG are only Xen; Proxmox is only KVM/containers; libvirtd does both). Then based on that you pick your actual host OS (in case it's not already set).
I have GPU passthrough of the iGPU on a Ryzen 4000U working with kvm/qemu+libvirtd on an Alpine base. So the hypervisor runs headless (booting in blind mode) and the guest OS takes over the iGPU entirely. The caveats are explained in that arch Wiki article, IIRC. I think the least obvious part for me was sourcing and loading of the microcode/firmware.
Also had success with hybrid graphics (Ryzen APU + AMD RX GPU with one passed through and one on host) on an otherwise similar hypervisor setup on Arch BTW.
Even more so than usual, the more recent and obscure your hardware, the higher chance you'll get to compare different kernel versions, incant random undocumented parameters you found in some forum comment, or even reach for out-of-tree patches to get running smoothly.
