Passing a physical GPU through to a Proxmox virtual machine can give the guest direct access to graphics hardware. It also changes how the host owns the device, which makes preparation and rollback more important than a one-line configuration tweak.
Check the platform before touching the VM
Confirm CPU virtualization and IOMMU support in firmware. Intel calls the latter VT-d; AMD systems typically expose AMD-Vi or IOMMU. Record the GPU model, its audio function and the firmware settings before making changes. Proxmox VE 8 and 9 may differ in bootloader and kernel handling; follow the documentation for the installed major release.
Inspect IOMMU grouping
Boot with IOMMU enabled and inspect kernel messages plus /sys/kernel/iommu_groups. A GPU must share an acceptable isolation group with the functions you intend to pass through. Do not assume that installing drivers inside the guest fixes poor hardware isolation; it does not.
Configure and test conservatively
Choose a VM machine type compatible with PCIe passthrough, add the PCI device in the Proxmox interface and attach associated audio functions when required. Keep another way to administer the host. If the host currently uses that GPU for its console, arrange an alternate console before binding it to vfio-pci.
Failure modes and rollback
A blank guest screen may be a guest-driver, firmware, reset or display-routing problem. Verify host device ownership and guest logs separately. Before rebooting after a change, write down how to revert kernel parameters and PCI bindings from an SSH session or rescue console.
A practical isolation review
Create a table listing every PCI function by vendor ID, device ID, driver and IOMMU group. Include the GPU's HDMI-audio component and any USB controller on the same card. For each group, record which function the hypervisor still needs. If an unrelated storage or network device sits in the same group, stop and assess platform isolation rather than assuming an ACS override makes the hardware secure. Test with an expendable VM first; GPU passthrough failures should not take down critical workloads.
Related reading
Continue with Proxmox maintenance.
Primary documentation: https://pve.proxmox.com/pve-docs/. Confirm the relevant release and hardware before applying changes to a production system.







