x86 GPU Passthrough with Qemu

Overview

GPU passthrough allows a virtual machine (VM) to use a dedicated GPU, enabling near-native performance for 3D workloads, media decoding/encoding and other GPU-intensive tasks.
QEMU combined with libvirt is a common stack used for this on Linux.

1. Requirements

Hardware

CPU with IOMMU support:
- Intel VT-d (Intel IOMMU)
- AMD-Vi (AMD)

2. Enable IOMMU in the Bootloader

Make sure following parameters added to kernel command line.
Terminal window
```
# For Intel CPUs
"intel_iommu=on"

# For AMD CPUs
"amd_iommu=on"
```

3. Identify the GPU

List devices
Terminal window
```
lspci -nn
```

4. Isolate GPU with VFIO

Blacklist the GPU driver on the host system so it won’t be used. Example
Terminal window
```
# For Intel GPUs
"module_blacklist=i915,xe"
```

Bind Devices to VFIO

options vfio-pci.ids=<VendorId>:<DeviceId>

5. Verify GPU Passthrough Works in QEMU

On Graphics VM (gui-vm) run the following commands.

To verifiy PCI device passthrough:
Terminal window
```
lspci | grep VGA
```
Check the driver:
Terminal window
```
lshw -c display
```

Limitation

With GPU passthrough to a specific VM, the GPU cannot be used for other VMs. However, sometimes we may need the GPU to be shared across different VMs, which is not possible with the current implementation.