CUDA version problem in container

[gyurobenjamin]

I'm running the linuxserver/faster-whisper:gpu container within my k8s cluster.
Container runtime is nvidia, GPU is correctly attached.

Error:
RuntimeError: CUDA failed with error CUDA driver version is insufficient for CUDA runtime version

Driver information on the host:
NVIDIA-SMI 550.144.03 Driver Version: 550.144.03 CUDA Version: 12.4

[lord0gnome]

I have the same issue:
| NVIDIA-SMI 570.124.04 Driver Version: 570.124.04 CUDA Version: 12.8 |
I updated cuda and my display drivers from version 550~ and cuda 12.4 and I still have the same error.

[MahmoudAshraf97]

What GPU are you using?

[lord0gnome]

I am using an A2000, but it turns out this had nothing to do with it, the graphics card was not correctly attached in my case. I was using podman play kube and didn't configure it properly.
In podman play kube it's

resources:
  requests:
    nvidia.com/gpu=0: 1

in a kubernetes manifest using the nvidia k8s plugin its

resources:
  requests:
    nvidia.com/gpu: 1

[gyurobenjamin]

What GPU are you using?

RTX 3090 FE

[kds69]

same here
RTX 3090 MSI

[gyurobenjamin]

same here RTX 3090 MSI

This worked for me: #1259 (comment)

Unsure why it requires these requests. Other services can access it without it.