Install NVIDIA® Drivers and CUDA® for RTX® 50 series

The launch of the RTX® 50 series was a major highlight of the first quarter of 2025. These new GPUs deliver substantially more power, giving users enhanced computing capabilities across many applications. Having thoroughly compared the RTX® 5090’s specs with its predecessor, the RTX® 4090, we’ll now examine how these new GPUs affect driver and NVIDIA® CUDA® Toolkit installation.

NVIDIA® Drivers

Ubuntu 22.04

RTX® 50 series GPUs require drivers version 570.xxx.xx or higher. Earlier versions aren’t supported. This means standard installation methods won’t work. To resolve this, you can use the NVIDIA® installer.

Start by installing some prerequisites:

sudo apt -y install gcc make

Next download the installer from the official website:

wget https://us.download.nvidia.com/XFree86/Linux-x86_64/570.124.04/NVIDIA-Linux-x86_64-570.124.04.run

Make the file executable:

sudo chmod +x NVIDIA-Linux-x86_64-570.124.04.run

Let’s start installing the driver:

sudo ./NVIDIA-Linux-x86_64-570.124.04.run

There is a critical issue to note: Selecting the proprietary driver during initial setup will appear to install correctly, but after rebooting, you’ll receive a “No devices were found” error. To avoid this, choose the MIT/GPL option instead, which will install the open source driver successfully.

After completing the installation, restart the server:

sudo shutdown -r now

Ubuntu 24.04

Currently, the ubuntu-drivers utility doesn’t recognize the RTX® 5090 or offer the latest driver installation. However, you can install it manually using two methods. The first method uses the operating system repository:

sudo apt install nvidia-driver-570-server-open

When installing the nvidia-driver-570-server package, you must include the open postfix. Without it, the system won’t be able to identify the GPU and will display the error message “No devices were found”.

After installing, you need to reboot the server using the command:

sudo shutdown -r now

The second method uses the installer, following the same process as Ubuntu 22.04.

NVIDIA® CUDA®

You’ll first need to choose which version of NVIDIA® CUDA® Toolkit to install. While PyTorch applications work the same on both versions 12.4 and 12.8, version 12.8 is recommended for advanced features and optimized machine learning algorithms since it fully supports the RTX® 50 series. Installation simply requires adding the NVIDIA® repository.

Download the special pin file:

wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2404/x86_64/cuda-ubuntu2404.pin

Put it into the working directory of the APT package manager:

sudo mv cuda-ubuntu2404.pin /etc/apt/preferences.d/cuda-repository-pin-600

Download the local copy of the NVIDIA® CUDA® Toolkit repository:

wget https://developer.download.nvidia.com/compute/cuda/12.8.1/local_installers/cuda-repo-ubuntu2404-12-8-local_12.8.1-570.124.06-1_amd64.deb

Let’s install the repository using the dpkg utility

sudo dpkg -i cuda-repo-ubuntu2404-12-8-local_12.8.1-570.124.06-1_amd64.deb

Add the GPG key to the Ubuntu keystore to ensure proper repository access:

sudo cp /var/cuda-repo-ubuntu2404-12-8-local/cuda-*-keyring.gpg /usr/share/keyrings/

Update packages cache and run the installation:

sudo apt-get update && sudo apt-get -y install cuda-toolkit-12-8

Don’t forget to reboot the server:

sudo shutdown -r now

See also:

• What is CUDA®
• Install NVIDIA® drivers in Linux
• Install CUDA® toolkit in Linux