You can also pass through more than just a GPU. I pass through an NVMe disk and a USB controller, effectively giving me a full second computer within my workstation. It's honestly fantastic.

Ok. And I assume that’ll all work at PCI speeds making it much more performant than using virtio-based forwarding. Sounds neat.

Do I need to do anything besides UEFI boot my virt-manager (QEMU) VM to make use of things like this?

I generally followed this guide, despite working from another distro: https://wiki.archlinux.org/title/PCI_passthrough_via_OVMF

The biggest pain every time I've done this is making sure that I can bind the device to the vfio driver.

Thanks.

So if I’m already UEFI booting my VMs with dedicated volumes, the key difference is how GPU pass-through is done?

Pretty much, lots of factors that are hardware dependent, such as how your motherboard assigns IOMMU groups come into play.