Found no NVIDIA driver on your system, but Driver is installed and working

[kodiakcreativity]

Before submitting an issue, please confirm it meets all the criteria

• I am on a distribution that is at least Ubuntu 20.04+, this means that your distros packages are the same as, or more up to date than Ubuntu 20.04, Or using Windows 10/11.

Describe the bug
A clear and concise description of what the bug is.
Recently updated NVIDIA GPU driver, mobile RTX 4080 to nvidia-driver-580 version 580.95.05-0ubuntu0.22.02.2.
Real video enhancer worked fine before updating the driver, was previously on version 570. other software recognizes and is able to use the GPU

To Reproduce
Steps to reproduce the behavior:
Update GPU driver message appears on startup:

ERROR DETECTED IN BACKEND SETUP!
ERROR: FATAL: Found no NVIDIA driver on your system. Please check that you have an NVIDIA GPU and installed a driver from http://www.nvidia.com/Download/index.aspx
ERROR: Found no NVIDIA driver on your system. Please check that you have an NVIDIA GPU and installed a driver from http://www.nvidia.com/Download/index.aspx
RVE Backend Version: 2.4.1-dev16
PyTorch Version: 2.6.0+cu126
PyTorch GPU 0: CPU
Half precision support: False
Available Backends: ['pytorch (cuda)']

Expected behavior
A clear and concise description of what you expected to happen.
This amazing application should work on this driver as it did on the previous.

Screenshots
If applicable, add screenshots to help explain your problem.

Logs
If the application generates a log, please upload it here.

frontend_log.txt and backend_log.txt

Desktop (please complete the following information):

• OS: [e.g. Ubuntu 20.04]
• Mint 21.3 Cinnamon. Kernel 6.8.0-90
• This is a notebook which can use intel emedded graphics or the discrete GPU, the error message seems to indicate it can see the intel embedded graphics but not the GPU, it worked fine before the driver update.

GPU
Please state your GPU, and what you are using to render (pytorch, tensorrt, ncnn)
NVIDIA RTX 4080

Additional context
Add any other context about the problem here.

[TNTwise]

Restart your computer, if you have it installed via flatpak, you need to update all flatpak apps too

[cliffbroadbent]

I can't say my experience is related, however it did start off with the same issue.

One day Ubuntu 24.04 was playing nice with RVE, the next it wasn't.

The story so far:

After running the usual apt-get update and apt-get upgrade, RVE is broken.

I experienced the same error as reported by @kodiakcreativity - even after a restart.

I use the binary found on the GitHub releases section.

The next chapter:

Thinking it may be a driver issue, I've installed Nvidia drivers 580 on Kubuntu 24.04 (and again on Kubuntu 25.10).

RVE now reports in the logs::

Using Device: NVIDIA GeForce RTX 3050 Ti Laptop GPU
Unable to load any of {libcudnn_graph.so.9.1.0, libcudnn_graph.so.9.1, libcudnn_graph.so.9, libcudnn_graph.so}
Invalid handle. Cannot load symbol cudnnCreate

This occurs with both RVE 2.3.4 and RVE 2.4.1

Details as reported by RVE 2.3.4:

System Information:
OS: Ubuntu 25.10 x86_64
CPU: AMD Ryzen 7 5800H with Radeon Graphics
RAM: 30.70 GB
Available Disk Space: 260.34GB

Software Information:
REAL Video Enhancer Version: 2.3.4
Python Version: 3.12.9
Utilizing local backend: False
Working Directory: /home/severin/.local/share/REAL-Video-Enhancer
Available Backends: ['tensorrt', 'pytorch (cuda)']
TensorRT Version: 10.9.0.34
PyTorch Version: 2.6.0+cu118
PyTorch GPU 0: NVIDIA GeForce RTX 3050 Ti Laptop GPU
Half precision support: True

backend_log.txt
frontend_log.txt

[TNTwise]

Deficiently looks like a driver issue. No issues on arch, maybe try uninstalling and reinstalling a different pytorch version?

[cliffbroadbent]

Thanks for the idea, however - no improvement.

Using RVE 2.3.4 that I know to have worked correctly, I have downloaded a fresh archive from the Releases section of GitHub.

This time, I selected pytorch 2.7.0.

When I attempt an upscale, I see the following - I note different versions of libs it's failing to load:

Using Device: NVIDIA GeForce RTX 3050 Ti Laptop GPU
Unable to load any of {libcudnn_graph.so.9.7.1, libcudnn_graph.so.9.7, libcudnn_graph.so.9, libcudnn_graph.so}
Invalid handle. Cannot load symbol cudnnCreate

Pytorch home screen report:

System Information:
OS: Ubuntu 25.10 x86_64
CPU: AMD Ryzen 7 5800H with Radeon Graphics
RAM: 30.70 GB
Available Disk Space: 270.55GB

Software Information:
REAL Video Enhancer Version: 2.3.4
Python Version: 3.12.9
Utilizing local backend: True
Working Directory: /home/severin/Documents/Real-Video-Enhancer/bin_2.3.4
Available Backends: ['pytorch (cuda)']
PyTorch Version: 2.7.0+cu128
PyTorch GPU 0: NVIDIA GeForce RTX 3050 Ti Laptop GPU
Half precision support: True

Output of nvidia-smi:

severin@boris:~$ nvidia-smi
Fri Jan 16 16:28:32 2026       
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 580.95.05              Driver Version: 580.95.05      CUDA Version: 13.0     |
+-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA GeForce RTX 3050 ...    Off |   00000000:01:00.0 Off |                  N/A |
| N/A   62C    P0             15W /   80W |     142MiB /   4096MiB |      2%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI              PID   Type   Process name                        GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|    0   N/A  N/A            1427      G   /usr/lib/xorg/Xorg                        4MiB |
|    0   N/A  N/A            2487      C   /usr/NX/bin/nxnode.bin                  119MiB |
+-----------------------------------------------------------------------------------------+

Output of cat /proc/driver/nvidia/version:

NVRM version: NVIDIA UNIX x86_64 Kernel Module  580.95.05  Tue Sep 23 10:11:16 UTC 2025

Are there any nvidia packages I should be installing outside of the current driver?

I can see the libraries it's complaining about are available, successfully downloaded by RVE when it first sets itself up:

./bin_2.3.4/python/python/lib/python3.12/site-packages/nvidia/cudnn/lib/libcudnn_graph.so.9

Logs as they stand after setup completes:
INSTALL_backend_log.txt
INSTALL_frontend_log.txt

[TNTwise]

Can you try torch version 2.6 then? it could be an issue introduced by newer torch version. I dont think this necessarily my code causing the issue.

[cliffbroadbent]

Summary:
I agree. I'm confident it isn't your code.

Looks like some permission or environment issue, however I can't diagnose any further.

Since that fateful apt-get upgrade, the cudnn issues have occured. "something" was changed by the Ubuntu package upgrade, however I haven't isolated what.

This has affected both an Ubuntu 24.04 and 25.10 installation.

Also, I wanted to say I really appreciate the skill and effort you've put into building RVE. Not to mention being so active on the issue tracker. Thank you.

The details:
The previous post was with torch 2.6, so can safely say I've tried both 2.6 and 2.7.

As a YOLO isolation test to investigate permissions as the culprit, I've run sudo ./REAL-Video-Enhancer and RVE works correctly.

I've checked permissions and all files under the RVE directory are owned by myself and group.

Can you provide me any hints for further investigation?

[TNTwise]

You could try the flatpak version