Nvidia-smi returns 'No devices were found' after installing driver

Centos 7
Nvidia driver 535.104.05
Cuda 11.7

When I entered ‘cat /sys/module/nvidia/version’, it returns 535.104.05.

I read several similar topics, and I know the reason may be the open source kernel, but I still have several questions:

  1. Do I have to do sudo reboot after installation?

  2. How can I do an uninstallation, and what version can I reinstall (the closed one)?

  3. Do I also need to uninstall Cuda?

The report.log is attached. Thank you for your patience.

nvidia-bug-report.log (1.0 MB)

Hi there @yikunhan42 and welcome to the NVIDIA developer forums.

Sep 14 05:47:59 node01 kernel: NVRM: Open nvidia.ko is only ready for use on Data Center GPUs.
Sep 14 05:47:59 node01 kernel: NVRM: To force use of Open nvidia.ko on other GPUs, see the
Sep 14 05:47:59 node01 kernel: NVRM: 'OpenRmEnableUnsupportedGpus' kernel module parameter described
Sep 14 05:47:59 node01 kernel: NVRM: in the README.

Yes, you do have the Open Kernel module installed. And the RTX 4090 is not a data center GPU.


It depends on what you used to install the driver. If you downloaded the .run file from NVIDIA you should use that as well to uninstall again. If you used yum than use that and use the equivalent of apt remove --purge. Also check in text-only mode if there might still be NVIDIA kernel modules loaded before doing the purge.
After that download the the default closed proprietary installer from NVIDIA or use the closed proprietarz driver recommended by your distribution package manager. I am not very familiar with CentOS so i can’t recommend how to find that exactly I am afraid.

If you installed CUDA manually as well, then yes to be safe. If you plan of using CUDA extensively I would also recommend to install the driver itself as part of the CUDA installation process. Please refer to the installation documentation. That way it is easier to avoid any kind of mismatches.

I hope that helps!