GPU passthrough in vSphere allows Windows virtual machines to access physical Nvidia GPUs directly, bypassing the hypervisor’s software rendering limitations. This process requires configuring PCIe passthrough on ESXi hosts and installing specific drivers within the guest OS. Proper setup ensures high-performance 3D acceleration for CAD, gaming, or video editing workloads without the overhead of virtual GPU sharing technologies.
Prerequisites and Hardware Compatibility
Before attempting to configure GPU passthrough, you must verify that your hardware supports Intel VT-d or AMD-Vi IOMMU technology. These virtualization extensions are mandatory for direct device assignment. Additionally, ensure that your Nvidia GPU is compatible with ESXi. While many consumer-grade cards can technically be passed through, official support and driver stability are significantly better with data center cards like the Nvidia Tesla or Quadro series designed for virtualization environments.
You should also check the CPU and motherboard BIOS settings. Ensure that Virtualization Technology (VT-x/AMD-V) and Interrupt Remapping are enabled in the BIOS. Without these settings active, the ESXi host will not expose the necessary IOMMU groups to the virtual machine, leading to immediate failure during the configuration process.
Disabling the Host GPU Driver
A critical step often overlooked is disabling the default Nvidia driver on the ESXi host itself. By default, ESXi may attempt to load a generic display driver or a vGPU manager if one is installed. If the host retains control of the GPU, the device cannot be assigned to a VM, resulting in boot failures or the device remaining unavailable in the vSphere Client.
To resolve this, you must remove the Nvidia vSphere driver package from the host. Log into the ESXi host via SSH or use the command line interface. Run the following command to list installed VIBs:
esxcli software vib list | grep nvidia
If you see packages related to
nvidia-vgpu or similar, uninstall them using:
esxcli software vib remove -n [package_name]
After removal, reboot the ESXi host. This ensures the GPU is completely released from the host’s management plane and is visible as an unassigned PCI device. This step is essential to prevent conflicts where the host tries to render graphics while the VM attempts to take exclusive control.
Configuring PCIe Passthrough in vSphere
Once the host is rebooted and the GPU is free, navigate to the vSphere Web Client. Select the ESXi host and go to the
Configure tab. Under the
Hardware section, click on
PCI Devices. You will see a list of all PCI devices attached to the host. Locate your Nvidia GPU in this list. It may be listed under a specific PCIe root port.
Select the GPU entry and check the box labeled
Passthrough. Click
OK to save the changes. The status of the device should change to indicate it is reserved for passthrough. Note that if the GPU is currently in use by any running VM or the host itself, you may need to migrate those workloads first.
Next, edit the settings of the target Windows virtual machine. Add a new
PCI Device adapter and select the Nvidia GPU from the dropdown menu. Ensure that the VM is powered off before making this change, as hot-plugging PCI devices is generally not supported for GPU passthrough scenarios. After adding the device, power on the VM.
Installing Drivers in the Guest OS
When the Windows VM boots, it will detect the Nvidia GPU as a new hardware device. Initially, it may install a basic Microsoft display driver, which does not provide 3D acceleration. To unlock full performance, you must install the standard Nvidia Windows drivers inside the VM.
Download the appropriate driver version from the Nvidia website. Avoid using the Data Center or GRID-specific drivers unless you have a valid license and specific requirement; standard Game Ready or Studio drivers often work well for general 3D acceleration tasks. Install the driver and reboot the Windows VM.
Upon reboot, open the Nvidia Control Panel or Device Manager to verify that the GPU is recognized and operating at its full clock speeds. You can run benchmark tools like Unigine Heaven to confirm that the VM is utilizing the physical hardware rather than falling back to software emulation.
Troubleshooting Common Issues
If the VM fails to boot after enabling passthrough, check the ESXi host logs (
/var/log/vmkernel.log) for IOMMU group errors. Sometimes, ACS overrides are needed for certain motherboards where PCIe switches do not properly isolate devices. However, applying ACS overrides is advanced and risky; always ensure the host GPU driver is disabled first, as mentioned earlier.
Another common issue is audio passthrough. If your Nvidia GPU has integrated HDMI/DisplayPort audio, you can add a second PCI device for the audio controller. This allows Windows to output sound through the GPU’s digital audio outputs, providing a complete multimedia experience.
For more information on managing virtual machine resources, see
Fixing Windows Boot Issues: chkdsk/r, Missing Drive C in BIOS, and NVIDIA Driver Installation. Additionally, if you are experiencing disk performance issues alongside GPU optimization, check
Reducing Disk Writes with Nvidia Drivers on Windows 11 Pro (24H2).
Frequently Asked Questions
Can I pass through multiple GPUs to a single VM?
Yes, vSphere supports passing through multiple PCI devices to a single virtual machine. You can add each GPU as a separate PCI adapter in the VM settings. However, ensure your application supports multi-GPU configurations, as most standard Windows applications only utilize one primary display adapter.
Does GPU passthrough affect VM snapshot capabilities?
Passing through physical GPUs can complicate snapshot creation. Some versions of vSphere may warn that snapshots are not supported or may fail when PCI passthrough devices are present. It is recommended to test snapshot functionality thoroughly in a non-production environment before relying on it for backup strategies.
Is SLI or NVLink supported with GPU passthrough?
Generally, no. Most consumer-grade SLI setups and professional NVLink bridges rely on direct communication between GPUs that is often broken when devices are isolated via IOMMU. Each GPU is typically treated as an independent entity by the VM, so multi-GPU scaling features may not function as they would on bare metal.
What happens if I forget to disable the host driver?
If the host driver remains active, the ESXi host retains ownership of the GPU. When you attempt to assign it to a VM, the operation will likely fail, or the VM will boot but show no graphics output because the host is still rendering to the display. Always reboot the host after removing the driver to ensure clean state release.
Are there performance penalties compared to bare metal?
GPU passthrough offers near-native performance, with minimal overhead compared to virtualized graphics solutions like VMware vGPU. The main penalty comes from the lack of direct hardware access to certain low-level APIs if the guest OS does not have proper drivers, but for standard DirectX and OpenGL workloads, the difference is negligible.