How to Manage vGPU on Vultr Cloud GPU Instances

Vultr Cloud GPU instances provide access to NVIDIA vGPU technology, enabling virtualized graphics processing for applications requiring GPU acceleration. This includes machine learning workloads, video processing, and virtual desktop infrastructure (VDI) solutions. Proper management of vGPU drivers and licensing ensures optimal performance and compliance with NVIDIA's licensing requirements.

Follow this guide to manage vGPU on your Vultr Cloud GPU instances.

Prerequisites

Before you begin, you need to:

• Have access to an Vultr Cloud GPU instance as a non-root user with sudo privileges.

Install DKMS Package

Dynamic Kernel Module Support (DKMS) is a Linux framework that simplifies the management of kernel modules by automatically building, installing, and updating them whenever the kernel changes. Kernel modules such as GPU drivers, VirtualBox drivers, or ZFS are normally compiled for a specific kernel version, which can cause compatibility issues after a system update. DKMS eliminates this problem by recompiling and reinstalling modules against the new kernel, ensuring they remain functional and consistent across upgrades without requiring manual intervention.

• Debian-based Distributions

  CONSOLE

  Copy

  $ sudo apt install -y dkms

  Explain Code

  Powered by Vultr AgentBeta

• RPM-based Distributions

  CONSOLE

  Copy

  $ sudo dnf install -y dkms

  Explain Code

  Powered by Vultr AgentBeta

Install NVIDIA Driver

Before installing, verify that your system detects the NVIDIA GPU hardware and confirm whether the NVIDIA kernel module is already loaded. The installation script should only be run if the GPU is detected but the driver is not installed.

Verify GPU and Kernel Module

1. Verify that your system detects the NVIDIA GPU hardware.

   CONSOLE

   Copy

   $ lspci | grep -i nvidia

   Explain Code

   Powered by Vultr AgentBeta

   This command should always return a result on GPU-enabled instances, even if drivers are not installed.

   Your output should be similar to the one below:

   06:00.0 3D controller: NVIDIA Corporation GA100 [A100 PCIe 80GB] (rev a1)

   If no output is displayed, your server does not have an attached NVIDIA GPU, and you should not proceed with driver installation.

2. Check if the NVIDIA kernel module is loaded.

   CONSOLE

   Copy

   $ lsmod | grep nvidia

   Explain Code

   Powered by Vultr AgentBeta

   Note

   If the above command shows an output, the NVIDIA driver is already installed and loaded; otherwise, it is not installed or not loaded, and you should proceed to the installation step below.

Driver Installation Steps

If your system detects the NVIDIA GPU but the driver is not loaded, you need to install it manually. Skip this step if the driver was already found in the previous check.

1. Ensure the DKMS package is installed, then run the NVIDIA driver installation script.

   CONSOLE

   Copy

   $ sudo /opt/nvidia/install.sh

   Explain Code

   Powered by Vultr AgentBeta

2. Reboot your system to load the new NVIDIA driver and nouveau blacklist to take effect.

   CONSOLE

   Copy

   $ sudo reboot

   Explain Code

   Powered by Vultr AgentBeta

3. After reboot, verify that the drivers are installed and the GPU is accessible.

   CONSOLE

   Copy

   $ lsmod | grep nvidia
   $ nvidia-smi

   Explain Code

   Powered by Vultr AgentBeta

   Output:

   nvidia_drm            122880  0
   nvidia_modeset       1355776  1 nvidia_drm
   ...
   Thu Aug 28 07:39:00 2025
   +-----------------------------------------------------------------------------------------+
   | NVIDIA-SMI 550.54.15              Driver Version: 550.54.15      CUDA Version: 12.4     |
   |-----------------------------------------+------------------------+----------------------+
   | GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
   | Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
   |                                         |                        |               MIG M. |
   |=========================================+========================+======================|
   |   0  NVIDIA A40-48Q                 On  |   00000000:06:00.0 Off |                    0 |
   ......

Update NVIDIA Driver

Before updating, verify that your system detects the NVIDIA GPU and that the current driver is installed. The update script should only be run if the GPU is present and the driver is already loaded.

1. Verify that your system detects the NVIDIA GPU hardware.

   CONSOLE

   Copy

   $ lspci | grep -i nvidia

   Explain Code

   Powered by Vultr AgentBeta

   If no output is displayed, your server does not have an attached NVIDIA GPU, and you should not proceed with updating the driver.

2. Check if the NVIDIA kernel module is loaded.

   CONSOLE

   Copy

   $ lsmod | grep nvidia

   Explain Code

   Powered by Vultr AgentBeta

   Note

   If this command shows no output, the driver is not installed, and you should complete the Install NVIDIA Driver section instead of updating. If output is displayed, proceed with the update.

Driver Update Steps

1. Run the NVIDIA driver update script.

   CONSOLE

   Copy

   $ sudo /opt/nvidia/update.sh

   Explain Code

   Powered by Vultr AgentBeta

   This updates the drivers while preserving your existing configuration and licensing settings.

2. Reboot your system to apply the update.

   CONSOLE

   Copy

   $ sudo reboot

   Explain Code

   Powered by Vultr AgentBeta

3. After reboot, verify that the updated drivers are active.

   CONSOLE

   Copy

   $ nvidia-smi

   Explain Code

   Powered by Vultr AgentBeta

   Your output should be similar to the one below:

   +-----------------------------------------------------------------------------------------+
   | NVIDIA-SMI 550.90.07              Driver Version: 550.90.07      CUDA Version: 12.4     |
   |-----------------------------------------+------------------------+----------------------+
   | GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
   ...

Troubleshoot Common Issues

If your NVIDIA drivers or vGPU services are not working as expected, use the steps below to diagnose and resolve common issues. These checks help you verify whether essential services are running, licensing is valid, and drivers are properly installed. If any component is missing or inactive, reinstalling the drivers with DKMS support usually resolves the problem.

Verify Driver Manager is Running

The nvidia-driver-manager.service unit file should be installed on your system as part of the NVIDIA driver package. This service is primarily used to manage GPU drivers but does not need to remain running continuously.

Check the driver manager service status.

$ sudo systemctl status nvidia-driver-manager.service

Output:

● nvidia-driver-manager.service - NVIDIA Driver Manager
     Loaded: loaded (/etc/systemd/system/nvidia-driver-manager.service; enabled; preset: enabled)
     Active: inactive (dead) since Tue 2025-08-12 19:15:11 UTC; 43s ago
   Main PID: 1060 (code=exited, status=0/SUCCESS)
        CPU: 63ms

If the service is missing, ensure your operating system's DKMS package is installed and reinstall the drivers using sudo /opt/nvidia/install.sh.

Check Licensing Status

The nvidia-gridd.service manages NVIDIA vGPU licensing to enable advanced features such as multi-instance GPU (MIG) and virtual desktop infrastructure (VDI). This service must be active and correctly configured for licensed vGPU deployments.

Verify that the service is running and license is present.

$ sudo systemctl status nvidia-gridd.service
$ sudo nvidia-smi -q | grep License

Output:

● nvidia-gridd.service - NVIDIA Grid Daemon
     Loaded: loaded (/usr/lib/systemd/system/nvidia-gridd.service; enabled; preset: enabled)
     Active: active (running) since Thu 2025-08-28 07:55:11 UTC; 1h 12min ago
   Main PID: 1945 (nvidia-gridd)
      Tasks: 3 (limit: 144799)
     Memory: 4.3M (peak: 5.0M)
        CPU: 25ms
     CGroup: /system.slice/nvidia-gridd.service
             └─1945 /usr/bin/nvidia-gridd
......
    vGPU Software Licensed Product
        License Status                    : Licensed (Expiry: 2025-8-29 7:55:13 GMT)

If nvidia-gridd.service is missing, ensure your OS DKMS package is installed for your operating system and reinstall the drivers using sudo /opt/nvidia/install.sh.

Check if Driver is Installed

Verify that the NVIDIA drivers are properly installed and loaded by checking both the driver status and the kernel modules.

$ sudo nvidia-smi
$ sudo lsmod | grep nvidia

Output:

Thu Aug 28 09:14:23 2025       
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.54.15              Driver Version: 550.54.15      CUDA Version: 12.4     |
|-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA A40-48Q                 On  |   00000000:06:00.0 Off |                    0 |
......
nvidia_uvm           4927488  0
nvidia_drm            122880  0
nvidia_modeset       1355776  1 nvidia_drm
......

If either nvidia-smi fails or no NVIDIA kernel modules are listed, ensure your operating system's DKMS package is installed and reinstall the drivers using sudo /opt/nvidia/install.sh.

Unsupported Linux Distributions

For Linux distributions that are not officially supported, you can use the alternative GPU installation script. This script provides best-effort support and may not receive the same level of validation as standard drivers.

Install Driver

1. Ensure that the DKMS package is installed on your system.

2. Run the alternative GPU installation script.

   CONSOLE

   Copy

   $ sudo bash /opt/nvidia/linux_gpu.sh

   Explain Code

   Powered by Vultr AgentBeta

   Important

   This script is provided as a courtesy and is best-effort only. Vultr does not provide support for custom installations or non-GPU deployments on GPU instances. Use at your own risk.

3. Reboot your system to apply the configuration and load the drivers

   CONSOLE

   Copy

   $ sudo reboot

   Explain Code

   Powered by Vultr AgentBeta

4. After reboot, verify the installation.

   CONSOLE

   Copy

   $ nvidia-smi

   Explain Code

   Powered by Vultr AgentBeta

   If successful, the output should confirm that the NVIDIA drivers are loaded and the GPU is accessible

Conclusion

Managing NVIDIA vGPU on Vultr Cloud GPU instances ensures that your GPU resources are properly configured, drivers are up to date, and licensing is valid. By following this guide, you have learned how to:

• Install and update NVIDIA drivers using DKMS for kernel compatibility.
• Verify GPU detection, driver status, and kernel modules.
• Check essential services such as nvidia-gridd.service and ensure licensing is active.
• Troubleshoot common issues and confirm proper driver installation.
• Use the alternative installation script for unsupported Linux distributions when necessary.