How to install the NVIDIA driver on Ubuntu 24.04 / 26.04 for a GPU server?

Updated: 28 September 2026 · Applies to: NVIDIA driver on Ubuntu 24.04 / 26.04 LTS (CUDA 13 needs driver 580 or newer)

Every GPU workload needs NVIDIA's driver first. On Ubuntu you can install it in two supported ways: with Ubuntu's own ubuntu-drivers tool (the simplest), or from NVIDIA's package repository. Use only one of the two methods on the same server. NVIDIA lists Ubuntu 22.04, 24.04 and 26.04 as supported.

Before you start

Log in over SSH as root or a user with sudo.
Update the system and restart if a new kernel was installed, so that the running kernel matches the installed kernel files:
```
sudo apt update && sudo apt upgrade -y
sudo reboot
```
Confirm the server sees the card: lspci | grep -i nvidia (see How to check that your GPU server can see its NVIDIA GPU?).

Method A: install with ubuntu-drivers (recommended)

Show the drivers that fit your card for compute workloads:
```
sudo ubuntu-drivers list --gpgpu
```
You get names such as nvidia-driver-535-server or nvidia-driver-550-open; the exact list depends on your card and Ubuntu release.
Let Ubuntu pick the right one:
```
sudo ubuntu-drivers install --gpgpu
```
To install one specific version from the list, add it to the command, for example sudo ubuntu-drivers install --gpgpu nvidia:535-server.
Restart the server:
```
sudo reboot
```
Log in again and check the driver:
```
nvidia-smi
```
A table with your GPU or GPUs means the driver works.

The packages that ubuntu-drivers installs are pre-built and signed, so they also work when Secure Boot is turned on.

Method B: install from NVIDIA's repository

Install the kernel files that the driver needs:
```
sudo apt install -y linux-headers-$(uname -r)
```

Add NVIDIA's repository. Use ubuntu2404 on Ubuntu 24.04 and ubuntu2604 on Ubuntu 26.04:

wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/x86_64/cuda-keyring_1.1-1_all.deb
sudo dpkg -i cuda-keyring_1.1-1_all.deb
sudo apt update

Install the driver with NVIDIA's open kernel modules:
```
sudo apt install -y nvidia-open
```
For a compute-only server without a desktop, NVIDIA's guide also lists sudo apt -V install libnvidia-compute nvidia-dkms-open.
Restart and check with nvidia-smi:
```
sudo reboot
```

Which driver version do I need?

CUDA 13 applications need driver 580 or newer. CUDA 12 applications need 525 or newer (12.8 needs 570 or newer).
Ollama needs driver 550 or newer for NVIDIA cards.
GPUs based on the Blackwell architecture (for example the RTX 5090) work only with NVIDIA's open kernel modules, that is the driver packages ending in -open. NVIDIA's open modules can be used on every Turing or newer GPU, which includes the A100, A40, RTX A5000, RTX 4090 and RTX 5090.

If something goes wrong

nvidia-smi says it could not communicate with the driver: see How to fix the "NVIDIA-SMI has failed" error (cannot communicate with the NVIDIA driver)?.
You want to switch from one method to the other: remove the old driver first with sudo apt purge '^nvidia-.*' '^libnvidia-.*' and sudo apt autoremove, restart, then follow the other method.

Frequently asked questions

Do I also need to install CUDA?
Only if you compile CUDA code yourself. PyTorch, Ollama and most AI tools only need the driver. If you do need the compiler, see How to install the CUDA Toolkit on Ubuntu 24.04 / 26.04?.

What does the "server" suffix mean in a driver name?
Ubuntu marks its Enterprise Ready Drivers with -server. They are meant for compute workloads. Drivers without the suffix are aimed at desktop and gaming use.

Can Ucartz install the driver for me?
Yes. Our engineers can install the driver, CUDA, PyTorch or a private LLM for you (see below).

Official documentation: Ubuntu Server documentation: NVIDIA drivers and NVIDIA driver installation guide for Ubuntu.

Need a GPU server, or a hand with the setup?

GPU dedicated servers: NVIDIA GPU servers for AI training and inference; our engineers can install the driver, CUDA, PyTorch or a private LLM and hand it over ready to use.
Private LLM installation: we install Ollama, the GPU driver and a chat interface on your own server.

Prefer a hand with the setup? Our engineers can do it for you: Hire an Expert, or use our on-demand server management.