The NVIDIA Driver was not detected. Driver 470.76 Cuda 11.4

I print:
docker run --name yolov4_csp -it -v /mnt/d/Datasets -v /mnt/c/Users/Alexander/Desktop/Mypetproject --shm-size=64g --gpus=all nvcr.io/nvidia/pytorch:21.05-py3 -gpu
and get this warning:
“WARNING: The NVIDIA Driver was not detected. GPU functionality will not be available.
Use ‘Nvidia-docker run’ to start this container; see
nvidia docker · NVIDIA/nvidia-docker Wiki · GitHub .”
±----------------------------------------------------------------------------+
| NVIDIA-SMI 470.28 Driver Version: 470.76 CUDA Version: 11.4 |
|-------------------------------±---------------------±---------------------+

windows 10 21390.2025
Ubuntu-18.04
docker-desktop 3.3.0

Is it because PyTorch isn’t compatible with the newest Cuda driver? Is there a way to downgrade the Cuda version if this is the reason?

Inside the container run ls -la /dev/dxg and if the device exists that means it can already access the GPU.

Those type of warnings appear because the containers still need to be updated, but the GPU should work. See point 7 in :: (nvidia.com)

Thank you for your answer!

1 Like