Isolate NVIDIA GPU for VFIO Passthrough (Arch Linux)

This guide helps you identify your NVIDIA GPU's PCI IDs, enable IOMMU, configure VFIO, and confirm the GPU is detached from the host OS.

---

Optional: Automate Binding/Unbinding with Python Script

Once you’ve successfully followed this guide, you can use a Python script to automatically bind or unbind your NVIDIA GPU from the host system with a simple command-line toggle.

➡️ View the script here

Make it executable and run with sudo:

chmod +x gpu-vfio-toggle.py
sudo ./gpu-vfio-toggle.py

This saves you from manually editing config files and regenerating initramfs every time.

---

Prerequisites:

Dual GPU Requirement

Your computer must have two GPUs: → a dedicated GPU (e.g. NVIDIA 4070) → and an integrated GPU (iGPU) (like Intel or AMD iGPU)

❗ Without both, GPU passthrough to a VM will not work — the host needs a separate GPU to stay usable.

Virtualization

Before starting, ensure your CPU supports virtualization and that it's enabled in your BIOS.

You must have:

• AMD-V (for AMD CPUs), or
• VT-x (for Intel CPUs)

Check if it's enabled:

Run:

lscpu | grep "Virtualization"

Expected output:

Virtualization:   AMD-V   # or "VT-x" for Intel

If you see nothing or it's blank:

• Reboot your machine
• Enter BIOS/UEFI (usually by pressing F2, Del, or Esc)
• Enable SVM Mode (for AMD) or Intel VT-x

---

Step 1: Identify PCI IDs

Run the following command to list NVIDIA devices with their PCI IDs:

lspci -nn | grep -E "NVIDIA"

Example output:

01:00.0 VGA compatible controller [0300]: NVIDIA Corporation AD104 [GeForce RTX 4070 SUPER] [10de:2783] (rev a1)
01:00.1 Audio device [0403]: NVIDIA Corporation AD104 High Definition Audio Controller [10de:22bc] (rev a1)

Extract the PCI IDs from the output. In this example:

10de:2783,10de:22bc

Copy and save these values into a text editor.

---

Step 2: Enable IOMMU

Edit GRUB config:

sudo nano /etc/default/grub

Find the line that starts with:

GRUB_CMDLINE_LINUX_DEFAULT="quiet splash"

Replace it with:

GRUB_CMDLINE_LINUX_DEFAULT="intel_iommu=on iommu=pt vfio-pci.ids=XXXX:XXXX,XXXX:XXXX"

XXXX:XXXX,XXXX:XXXX = Your PCI IDs

Remove quiet splash if present.

For AMD users, use amd_iommu=on instead of intel_iommu=on.

---

Reload GRUB:

sudo grub-mkconfig -o /boot/grub/grub.cfg

---

Reboot:

sudo reboot

---

Step 3: Isolate GPU from Host

Create VFIO config file:

sudo touch /etc/modprobe.d/vfio.conf
sudo nano /etc/modprobe.d/vfio.conf

Add the following content (replace IDs if needed):

options vfio-pci ids=10de:2783,10de:22bc
softdep nvidia pre: vfio-pci

---

Regenerate initramfs:

Arch

sudo mkinitcpio -p linux

For other non-Arch LinuxOS users, use sudo update-initramfs -c -k $(uname -r)

---

Reboot again:

sudo reboot

---

Step 4: Confirm VFIO Driver is in Use

Run the following to check kernel drivers in use:

lspci -k | grep -E "vfio-pci|NVIDIA"

Expected output snippet:

Kernel driver in use: vfio-pci
Kernel driver in use: vfio-pci