Configuring the Proxmox hosts

Access your Proxmox host via SSH or the web GUI’s Shell option.

Step 1: Enable IOMMU and Intel GVT-g

#

Edit the GRUB configuration file /etc/default/grub and update the GRUB_CMDLINE_LINUX_DEFAULT variable to:

GRUB_CMDLINE_LINUX_DEFAULT="intel_iommu=on iommu=pt i915.enable_gvt=1 pcie_acs_override=downstream,multifunction"

This is why you need this: intel_iommu=on iommu=pt i915.enable_gvt=1 enables the core trio of IOMMU, passthrough, and GVT-g (mediated devices) functionality. The pcie_acs_override=downstream,multifunction parameter does the heavy lifting by forcing hardware into separate IOMMU groups. No more dealing with device clusters – you get individual control over each piece of hardware, plus the sweet benefit of a stable, non-crashing host.

Apply these new boot parameters by running:

sudo proxmox-boot-tool refresh

Step 2: Load Required Kernel Modules

#

Those parameters won’t work their magic without the right kernel modules loaded first. Add these to /etc/modules:

# Modules required for PCI passthrough
vfio
vfio_iommu_type1
vfio_pci
vfio_virqfd

# Modules required for Intel GVT-g Split
kvmgt
vfio_mdev
i915

Step 3: Update Boot Image

#

Regenerate the boot image to include the new modules:

sudo update-initramfs -u -k all

Step 4: Reboot

#

Restart your Proxmox host for all changes to take effect:

sudo reboot

While rebooting, make sure you enabled all virtualization options in your BIOS like described here.

Step 5: Verify Configuration After Reboot

#

After reboot, verify the devices are properly separated into different IOMMU groups. Replace <node> with your actual PVE node name:

sudo pvesh get /nodes/<node>/hardware/pci --pci-class-blacklist ""

Your output should look similar to this:

┌──────────┬────────┬──────────────┬────────────┬────────┬──────────────────────────────────────
│ class    │ device │ id           │ iommugroup │ vendor │ device_name
╞══════════╪════════╪══════════════╪════════════╪════════╪══════════════════════════════════════
...
│ 0x020000 │ 0x154d │ 0000:01:00.0 │         10 │ 0x8086 │ Ethernet 10G 2P X520 Adapter
├──────────┼────────┼──────────────┼────────────┼────────┼──────────────────────────────────────
│ 0x020000 │ 0x154d │ 0000:01:00.1 │         11 │ 0x8086 │ Ethernet 10G 2P X520 Adapter
├──────────┼────────┼──────────────┼────────────┼────────┼──────────────────────────────────────
│ 0x030000 │ 0x3e92 │ 0000:00:02.0 │          0 │ 0x8086 │ CoffeeLake-S GT2 [UHD Graphics 630]
├──────────┼────────┼──────────────┼────────────┼────────┼──────────────────────────────────────
│ 0x040380 │ 0xa348 │ 0000:00:1f.3 │          9 │ 0x8086 │ Cannon Lake PCH cAVS
...

Take note of the GPU ID (which is 0000:00:02.0 in this example), as you’ll use it in the next step.

Step 6: Confirm GVT-g Support

#

Verify that GVT-g split is working and mediated devices are available. Replace <gpu id> with your GPU’s ID from the previous step:

sudo ls /sys/bus/pci/devices/<gpu id>/mdev_supported_types

You should see the following output:

i915-GVTg_V5_4  i915-GVTg_V5_8

Step 7: Configure Additional Hosts

#

If you have multiple Proxmox VE hosts with GPUs, repeat these steps on each host.

TIP
If you have a Proxmox VE cluster, you can create a resource mapping (Datacenter → Resource Mappings). This mapping can then be added to your VM hardware instead of a directly connected raw device.