GPU drivers

RemoteGPU NVIDIA images include a driver when you create or rebuild a GPU VM. You decide when to update or replace that driver. If you change an existing CPU VM to a GPU plan, the VM keeps its disk and software; you may need to install a driver yourself.

Verify GPU access

Connect to your VM through SSH or the browser console, then run:

nvidia-smi --query-gpu=name,driver_version,memory.total --format=csv

The output lists each GPU's name, driver version, and total memory. Check that the number of GPUs matches your plan, then try your application. Keep the installed driver unless the application needs a different version.

If the command is missing or reports an error, use Troubleshoot driver results before changing drivers. For more GPU queries, see the NVIDIA-SMI reference.

Install a driver on Ubuntu

Use these steps after confirming that the VM has an NVIDIA GPU and needs a driver installation. Save your work first: you may need to reboot. If a driver is already installed, check how it was installed before replacing it.

Check the Ubuntu release and kernel running in your VM:

cat /etc/os-release
uname -r

The first command identifies your Ubuntu release; the second prints the running kernel version. Check Ubuntu's NVIDIA driver guide for the requirements that apply to your system before installing.

1. If ubuntu-drivers is missing, install the tool:

   sudo apt update
   sudo apt install ubuntu-drivers-common

2. List the drivers Ubuntu offers for GPU compute:

   ubuntu-drivers list --gpgpu

3. Let Ubuntu select a driver for the detected hardware:

   sudo ubuntu-drivers install --gpgpu

Follow Ubuntu's guide for the matching driver utilities and, if your GPU configuration requires it, the software for NVSwitch connections between GPUs. Complete the guide's reboot steps, then repeat the GPU access check above.

Installation can replace drivers already on the VM. If it fails, some packages may have changed while others are unfinished. Read and resolve the error before trying another installation method; the previous setup is not restored for you.

Troubleshoot driver results

Use the result to choose the next check:

Result	Meaning	Next check
nvidia-smi command not found	Your shell cannot find the NVIDIA utility	Check whether the utility was installed and where it is located
Driver communication error	The utility cannot connect to the driver	Check whether the driver loaded, supports the running kernel, and installed successfully
Driver/library version mismatch	The running driver and installed libraries have different versions	Check whether a driver update needs a reboot
No NVIDIA devices in lspci	The VM cannot see any NVIDIA hardware	Confirm the GPU plan change finished; contact support if the GPU is still missing

After a driver update, a reboot may be needed to load the new driver and clear a version mismatch. See Ubuntu's troubleshooting guidance.

Check assigned hardware

If the driver cannot find a GPU, check which NVIDIA devices the VM can see:

lspci -nn -d 10de:

If no devices appear, confirm that the GPU plan change completed and contact support. If lspci is missing, report that result to support instead.

Check an existing driver

Check the loaded NVIDIA driver version:

cat /proc/driver/nvidia/version

On Ubuntu, this command lists matching driver packages with their versions and installation status:

dpkg-query -W -f='${Package}\t${Version}\t${Status}\n' \
  'nvidia-driver-*' 'nvidia-dkms-*' 'nvidia-utils-*' \
  'linux-modules-nvidia-*' 'libnvidia-compute-*'

If the version file is missing, the driver may not be loaded. If the package list is empty, a driver may have been installed another way. Check how the VM was set up before installing a replacement.

Read next

• Read Manage a VM to change its compute plan.
• Read VM troubleshooting if the VM cannot start or you cannot log in.