A Comprehensive Guide to GPU Passthrough with KVM on Linux
Virtual machines are a cornerstone of modern computing, offering incredible flexibility for testing, development, and system isolation. However, they have historically shared one significant weakness: poor graphics performance. Standard virtualized GPUs are sufficient for basic desktop tasks, but they fall short for anything demanding, like gaming, 3D rendering, or machine learning.
This is where GPU passthrough comes in. By using a technology called VFIO (Virtual Function I/O), you can grant a virtual machine exclusive, direct control over a physical GPU installed in your host machine. The result is near-native graphics performance inside your VM, unlocking capabilities that were once impossible.
This guide will walk you through the concepts, requirements, and steps needed to configure GPU passthrough on a Linux host using KVM (Kernel-based Virtual Machine).
Why Bother with GPU Passthrough? The Benefits
Assigning a dedicated graphics card to a virtual machine is a complex process, but the payoff is substantial.
- High-Performance Gaming: Run demanding Windows games inside a Linux-hosted VM without the need to dual-boot. You can enjoy the stability and power of a Linux workflow while accessing the vast Windows gaming library at full speed.
- Professional Applications: Utilize GPU-accelerated software for video editing, CAD, 3D modeling, and scientific computing within a secure, isolated virtual environment.
- Hardware Consolidation: Run multiple operating systems simultaneously on a single powerful machine, each with dedicated hardware resources, reducing the need for separate physical computers.
- Enhanced Security: By isolating a GPU-intensive task to a VM, you can protect your host operating system from potentially unstable drivers or software.
The Essential Checklist: Hardware and Software Requirements
Before you begin, it’s crucial to ensure your hardware is compatible. GPU passthrough relies on specific CPU and motherboard features to function correctly.
Hardware Requirements:
- A CPU with Virtualization Support (IOMMU): This is the most critical requirement. Your processor must support IOMMU (Input-Output Memory Management Unit). This feature allows the host to manage and direct hardware access for virtual machines.
- For Intel CPUs, this feature is called Intel VT-d.
- For AMD CPUs, it is called AMD-Vi.
- A Motherboard with IOMMU Support: Your motherboard’s BIOS/UEFI must have an option to enable VT-d or AMD-Vi. Most modern motherboards support this, but it’s essential to verify.
- Two Graphics Cards: You need one GPU for your host Linux system and a separate, dedicated GPU to pass through to the virtual machine. An integrated GPU (like Intel HD Graphics) can serve as the host GPU, leaving a powerful dedicated card for the VM.
- Sufficient System Resources: Ensure you have enough RAM and CPU cores to comfortably run both the host OS and the resource-intensive guest VM.
Software Requirements:
- A modern Linux distribution (e.g., Arch Linux, Fedora, Ubuntu).
- KVM, QEMU, and
libvirtinstalled. - A graphical management tool like
virt-manageris highly recommended for easier configuration.
The Step-by-Step Process: Configuring Your System
Once you’ve confirmed your hardware is compatible, you can begin the software configuration. This process involves preparing the host system to release control of the GPU so the VM can claim it.
Step 1: Enable Virtualization in Your BIOS/UEFI
Reboot your computer and enter the BIOS/UEFI setup. Navigate to the CPU or chipset settings and enable both virtualization (VT-x/SVM) and the IOMMU feature (Intel VT-d/AMD-Vi). Save your settings and reboot.
Step 2: Configure the Linux Host Kernel
You must instruct the Linux kernel to enable IOMMU and prepare the target GPU for passthrough. This is typically done by editing your bootloader’s configuration file (e.g., /etc/default/grub).
You will need to add parameters to the kernel command line, such as intel_iommu=on or amd_iommu=on. You also need to identify the Vendor and Device IDs of the GPU you intend to pass through and assign it to the VFIO driver using a parameter like vfio-pci.ids=10de:1f08,10de:10f9 (these IDs are just an example). After editing the file, you must update your bootloader configuration and reboot.
Step 3: Verify IOMMU Groups and Isolate the GPU
IOMMU organizes devices into groups. For a successful passthrough, the target GPU should ideally be in its own isolated group. You can run a script or command to list your IOMMU groups and verify the GPU (and its associated components, like its HDMI audio device) are properly grouped and ready for passthrough. This step ensures that when you pass the GPU to the VM, you aren’t forced to pass through other essential system devices along with it.
Once configured correctly, the vfio-pci kernel driver will claim the GPU at boot, preventing the host Linux system from using it and making it available for virtual machines.
Step 4: Create and Configure the Virtual Machine
Using virt-manager or by directly editing the VM’s XML configuration, you can now add the hardware.
- Create a new virtual machine, but do not install the OS yet.
- In the VM’s hardware details, remove the default emulated display (like Spice or QXL).
- Click “Add Hardware” and select “PCI Host Device”.
- Find your dedicated GPU and its associated audio device in the list and add them to the VM.
- Proceed with installing your guest operating system (e.g., Windows). Once inside the VM, you can install the official graphics drivers just as you would on a physical machine.
Common Pitfalls and Troubleshooting
GPU passthrough can be challenging, and several common issues may arise.
- The “Code 43” Error in Windows: NVIDIA drivers are known to detect when they are running in a virtual environment and may disable the card, showing a “Code 43” error in the Device Manager. Hiding the KVM hypervisor signature from the guest is a common and effective workaround for this issue.
- Poor IOMMU Grouping: If your GPU is in the same IOMMU group as other critical hardware (like your primary USB controller), you may not be able to isolate it cleanly. An advanced (and potentially system-destabilizing) solution is to use an ACS override patch for your kernel, but this should be considered a last resort.
- Black Screen on VM Start: If the VM boots to a black screen, it could be due to several factors, including an incorrectly passed VBIOS, driver conflicts on the host, or an improper VM configuration.
A Final Word on Security
While powerful, GPU passthrough fundamentally alters the security model of virtualization. By giving a virtual machine direct, low-level access to a piece of hardware, you are bypassing some of the security layers that typically isolate a VM.
Because of this, it is critical to only run trusted operating systems and applications inside a VM with a passthrough device. A compromised guest OS could theoretically exploit this direct hardware access to attack the host system.
Setting up GPU passthrough is a project for the determined enthusiast, but the reward is a uniquely powerful and flexible system that combines the best of all worlds.
Source: https://infotechys.com/gpu-passthrough-on-kvm/
