Skip to content

NVIDIA Drivers

From TensorFlow 2.14 onward, only the NVIDIA driver is required — you no longer need to install CUDA or cuDNN manually. This page explains how to check your current driver and install or update it if needed.

Check your current driver

nvidia-smi

If a driver is installed, you will see a table similar to this:

+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 535.183.01   Driver Version: 535.183.01   CUDA Version: 12.2              |
|-----------------------------------------+----------------------+----------------------+
|   0  NVIDIA RTX A3000 12GB          Off | 00000000:01:00.0 Off |                  Off |
|  N/A   50C    P0              N/A / 90W |      8MiB / 12288MiB |      0%      Default |
+-----------------------------------------+----------------------+----------------------+

Note

The CUDA version shown is the maximum supported by your driver, not an installed version. You do not need to install CUDA separately.

If nvidia-smi is not found or the driver is outdated, follow the steps below.

Install or update the driver

Update package lists

sudo apt update && sudo apt upgrade

List compatible drivers for your GPU

sudo ubuntu-drivers list

You will see output like:

nvidia-driver-535
nvidia-driver-545
nvidia-driver-570
...

Install the latest driver

Pick the highest non-open non-server version from the list:

sudo apt install nvidia-driver-570
sudo reboot

Verify

After rebooting, run nvidia-smi again. The new driver version should be shown.

Minimum driver version

The driver must support a CUDA version compatible with TensorFlow. You can always check the required CUDA version for the TensorFlow release used by IGM in setup.py, then cross-reference it with the TensorFlow GPU build table and the CUDA–driver compatibility table.

As a rule of thumb, driver version ≥ 525 supports CUDA 12.x and is sufficient for all recent IGM releases.