CUDA error messages on Docker/WSL installed TF

TensorFlow
,
1

I have a Windows 11 Pro laptop, and I use WSL and Docker images to run Tensorflow. Lately, I notice CUDA related error messages when I do just “import tensorflow as tf”. But the program seems to still work okay.

My Windows, WSL, and Docker are up to date. I issue the following command in WSL to run TF in Jupyter notebook:

docker run -it --rm -p 8888:8888 --gpus all tensorflow/tensorflow:latest-gpu-jupyter

I received 3 error messages when I just ran “import tensorflow as tf” in the Jupyter notebook. They are related to unable to register CuFFT, CuDNN, and CuBLAS because they are already registered.

What do I do to get rid of these error messages? The following is a screenshot from the Jupyter notebook.

image
image2459×949 232 KB

3

Hi @kdlin1, This is a known issue. You can refer to this issue to know the further updates. I think those are warnings, even though you are getting those warnings but still you can able to detect the GPU. Thank You.

5

Thanks for the link. That link has info that 2.16.1 has no such issue. So I did the following:

docker run -it --rm -p 8888:8888 --gpus all tensorflow/tensorflow:2.16.1-gpu-jupyter

It does not have error messages anymore.

7

Hi @kdlin1, Yes you are right the issue is not replicating in the docker I have also tested it, it was replicating on the local environment while importing Tensorflow. Thank You.