How to Set Up GPU Passthrough on KVM for a High‑Performance Linux VM

GPU passthrough on KVM

A detailed, step‑by‑step guide to setting up GPU passthrough on KVM for Linux hosts. Learn how to enable IOMMU, bind your GPU to VFIO, configure hosts and guests, avoid common issues, and unlock near‑native GPU performance in your virtual machine.

Table of Contents

🔈Introduction

If you want to squeeze out every bit of performance in a virtual machine—whether for gaming, 3D rendering, video editing, machine learning, or other GPU‑intensive work—GPU passthrough on KVM/VFIO is one of your best bets. This guide gives you a step‑by‑step, practical walkthrough (with commands, tables, and tips) to get GPU passthrough working correctly. I’ve designed it so you can follow it reliably, avoid common pitfalls, and achieve near‑native GPU performance in your guest VM.

✅ What is GPU Passthrough (and Why Use It)

GPU passthrough is a technique by which you dedicate a physical GPU (and its associated devices, like audio over HDMI/DisplayPort) to a virtual machine so that the guest OS can drive it directly. The result: the guest gets almost native performance since the virtualization overhead is minimized.

Benefits include:

  • Full utilization of GPU for compute / rendering
  • Better performance than virtualized GPU drivers or paravirtualized 3D acceleration.
  • Flexibility: run multiple OSes, isolate workloads, etc.

Key drawbacks / requirements:

  • Need hardware support (CPU, motherboard, BIOS/UEFI) for IOMMU / VT-d / AMD-Vi
  • Need two GPUs (or at least one discrete GPU for host + one for guest) or the ability to use integrated graphics for host
  • Complex configuration; small mistakes can lead to black screen, guest doesn’t boot, driver issues, etc.

🧰 Hardware & Host Requirements

Here are the minimum and recommended hardware/software requirements:

ComponentMinimum RequirementRecommended / Best Practice
CPUSupports Intel VT‑d (for Intel) or AMD‑V/AMD‑Vi / IOMMULatest generation with good IOMMU grouping, hardware interrupt remapping
Motherboard / BIOS/UEFIBIOS that can enable VT‑d / IOMMU, secure boot (optional)UEFI firmware, good PCIe passthrough support, ACS (Access Control Services) if required
GPU(s)One discrete GPU for host, another to passthrough to guest, or using iGPU for hostTwo discrete GPUs (one modest for host), or iGPU + discrete GPU combo; vendor drivers that work under VFIO
MemoryEnough RAM to be shared between host + VM comfortablyPlenty of RAM (e.g. 32 GB+), with support for hugepages
Linux HostRecent kernel (5.x or newer), support for VFIO, libvirt / QEMU, OVMF / UEFI for guest if neededStable distro, up‑to‑date kernel, virt‑manager / virsh, good AMD/NVIDIA driver support (if needed)

▶️ GPU passthrough on KVM: Step‑by‑Step Setup Guide

🔄 Step 1: Enable Virtualization & IOMMU in BIOS/UEFI

  • Reboot into BIOS/UEFI.
  • Enable VT‑x / VT‑d (Intel) or AMD‑V / AMD‑Vi / IOMMU (AMD)
  • If there are options for PCIe ACS or interrupt remapping, enable them
  • Save and reboot

🔄 Step 2: Configure Kernel Boot Parameters

You need to pass the correct parameters so the host kernel enables IOMMU and VFIO early. Start with editing the /etc/default/grub file:

				
					sudo vim /etc/default/grub

Find the line starting with GRUB_CMDLINE_LINUX_DEFAULT (or analogous). Append these parameters:

  • For Intel CPUs
				
					intel_iommu=on iommu=pt
vfio-pci.ids=XXXX:YYYY,AAAA:BBBB
  • For AMD CPUs:
				
					amd_iommu=on iommu=pt
vfio-pci.ids=XXXX:YYYY,AAAA:BBBB

Replace XXXX:YYYY etc. with the vendor:device IDs of your GPU + GPU audio device (you’ll get those in the next step). iommu=pt (pass‑through mode) helps reduce overhead. Then update GRUB and reboot:

				
					sudo update-grub                                  # Debian/Ubuntu
				
			
				
					sudo grub2-mkconfig -o /boot/grub2/grub.cfg       # for RHEL/CentOS/AlmaLinux/Fedora
				
			
				
					sudo reboot

🔄 Step 3: Identify the GPU & IOMMU Groups

You need to find out which PCI devices correspond to the GPU, its associated audio device, and which IOMMU group they are in.

				
					# List PCI devices with numeric IDs
lspci -nn
				
			
				
					# For grouping:
for g in /sys/kernel/iommu_groups/*/devices/*; do
    grp=${g#*/iommu_groups/*}; grp=${grp%%/*}
    echo "IOMMU Group $grp:"
    lspci -nns ${g##*/} | sed 's/^/\t/'
done

You’ll see output listing IOMMU groups. Example:

				
					IOMMU Group 3:
    01:00.0 VGA compatible controller [0300]: NVIDIA Corporation RTX 3080 [10de:2206]
    01:00.1 Audio device [0403]: NVIDIA Corporation Device [10de:1aef]
💡Make note of all devices in the same group—you’ll need to passthrough all of them (audio + GPU) for correct behavior. If other unrelated devices are in the same group, you may need to use ACS patch or accept those extra devices being passed through.

🔄 Step 4: Prevent Host from Loading GPU Drivers / Bind the GPU to VFIO

You need to prevent the host from using the GPU. That means blocking or disabling default drivers (e.g. nouveau, nvidia, amdgpu) and binding the device to vfio-pci.

✅ Example: Create vfio config

Create a file /etc/modprobe.d/vfio.conf:

				
					sudo vim /etc/modprobe.d/vfio.conf

Add lines like:

				
					options vfio-pci ids=10de:2206,10de:1aef
✋NOTE: Replace with your GPU vendor:device IDs, including its audio function.

Also disable conflicting drivers (e.g. for NVIDIA):

				
					sudo vim /etc/modprobe.d/disable‑nvidia.conf

Add the following lines:

GPU passthrough on KVM

Photo by admingeek from Infotechys

✅ Update initramfs, reboot

				
					sudo update-initramfs -u                        # Debian / Ubuntu
				
			
				
					sudo dracut -f --kver $(uname -r)               # Fedora / RHEL / AlmaLinux
				
			
				
					sudo reboot

🔄 Step 5: Verify Binding

After reboot, check that the GPU is using vfio-pci:

				
					lspci -nnk -d <GPU_VENDOR>:<GPU_DEVICE>

Example output should show something like:

				
					01:00.0 VGA compatible controller [0300]: NVIDIA Corporation RTX 3080 [10de:2206]
    Subsystem: ...
    Kernel driver in use: vfio-pci
    Kernel modules: nvidia, nouveau, vfio_pci

Also check dmesg for IOMMU:

				
					dmesg | grep -e DMAR -e IOMMU

Ensure that there are lines indicating IOMMU is active.

🔄 Step 6: Create / Configure VM with GPU Passthrough

Now it’s time to set up or modify your VM so that it uses this GPU.

🔹Using virt‑manager (GUI)

  • Create a new VM or edit an existing one
  • Under CPU settings, set CPU model to host-passthrough (this ensures the guest sees the host CPU directly).
  • Under Firmware choose UEFI/OVMF (better for modern OSes, secure boot, etc.)
  • Under Add Hardware → PCI Host Device, add the GPU device and its audio function (both).
  • For video/display, disable or remove any virtual GPU devices (like QXL), since you’re using the physical GPU.
  • Configure disk: use virtio or virtio‑scsi for better disk I/O. Cache options like cache=none or io=threads help. See tuning section below.
GPU passthrough on KVM

Photo by admingeek from Infotechys

🔹Using XML (virsh edit <vm>) / CLI

You may want to edit the VM’s XML to include something like:

				
					<cpu mode='host-passthrough' check='partial'/>
<features>
    <kvm>
        <hidden state='on'/>
    </kvm>
    <hyperv>
        <relaxed state='on'/>
        <vapic state='on'/>
        <spinlocks state='on' retries='8191'/>
        <vendor_id state='on' value='YOUR‑VENDOR′ID'/>
    </hyperv>
</features>

<devices>
  <hostdev mode='subsystem' type='pci' managed='yes'>
    <source>
       <address domain='0x0000' bus='0x01' slot='0x00' function='0x0'/>
    </source>
    <rom file='/usr/share/qemu/gpu‑dump.rom'/>
  </hostdev>
  <hostdev mode='subsystem' type='pci' managed='yes'>
    <source>
       <address domain='0x0000' bus='0x01' slot='0x00' function='0x1'/>
    </source>
  </hostdev>
</devices>

The ROM file line is optional but helps with some NVIDIA cards or where you see instability / driver issues.

🔄 Step 7: Guest OS Installation & Drivers

  • Boot the guest OS (e.g. Windows 10 / 11 or a Linux distro)
  • Install GPU drivers as if it were native hardware
  • If Windows shows “Code 43” (NVIDIA error) or detects it’s virtualized, the CPU config / hidden hypervisor flags might help workaround that. Using host-passthrough CPU + hiding hypervisor helps.

🔧 Performance Tuning & Tweaks

AreaWhat to TuneExample / Reason
Memory BackingUse hugepages or hugeTLB so that guest memory is backed more efficientlyAdd <memoryBacking><hugepages/></memoryBacking> to VM XML; reserve hugepages on host temporarily for the VM.
Disk I/OUse virtio‑scsi or virtio‑blk with cache=none and multiple queuesIn XML or via virt‑manager, pick virtio‑scsi, set appropriate io threads and queue count. 
CPU PinningPin vCPUs to specific physical cores for consistency and less resource contentionUse virsh vcpupin or define in XML.
GPU ROM / Vendor ID tweaksFor NVIDIA or other cards, using dumped ROM or setting custom vendor_id avoids driver detection issuesThe ROM file line in hostdev section; patching/passing vendor_id in hyperv features. 

🧰 Troubleshooting Common Issues

SymptomPossible CauseSolution
Black screen / guest won’t boot with GPU passedGPU still bound to host or BIOS/UEFI settings wrongCheck lspci -nnk, check driver binding; ensure BIOS settings correct
Guest shows low resolution or no display outputFirmware mismatch (BIOS vs UEFI), wrong GPU ROM, missing audio deviceUse OVMF, correct ROM, include all hostdev functions (audio + graphics)
Windows Code 43 / driver blockingVirtualization detection, CPU model not host‑passthroughAdd hyperv features, disable hypervisor bit, set CPU model to host‑passthrough
IOMMU group includes unwanted devicesShared group, motherboard limitationIf possible, enable ACS or use motherboard with better isolation; else pass all devices or accept the grouping

📌 Conclusion

GPU passthrough with KVM is a powerful way to give a virtual machine almost native graphics performance. The steps are technical, and there are many places where things can go wrong, but if you follow through in a structured way—confirming hardware support, identifying devices properly, binding drivers, setting up the VM with the right CPU and firmware options—you’ll arrive at a stable, high‑performance guest.

Did you find this article helpful? Your feedback is invaluable to us! Feel free to share this post with those who may benefit, and let us know your thoughts in the comments section below.

📕 Related Posts
RHEL9 on KVM
Commands
Install RHEL9 on KVM

In this tutorial, we’ll guide you through the step-by-step installation of RHEL9 on KVM. The process is uncomplicated, and we’ll present it in a format

Read More »
Tags

Leave a Reply

Your email address will not be published. Required fields are marked *