Tensorflow and CUDA version

Hailo DFC is not recognizing the GPU. I noticed that it uses TensorFlow 2.12, while I have CUDA 12.6. Should I downgrade my CUDA version? Any recommendations?

I recommend to use the Hailo AI Software Suite docker. It contains all the dependencies. Check the script used to start the Docker. It will show you what tests are run before GPU support is enabled.

Here are some post that have some additional information:

Hailo Community - GPU not detected in docker container

Hailo Community - GPU not detected in Hailo Suite

I had the same issue, but I have had my fair share headaches with TF. You can install all the tools using this commit of their docs.

After that, the optimization should see the GPU. My optimization time went from 10 min to 1 minute using 88 cores to 1 RTX 2080Ti.

Also make sure you have installed CUDA 11.8 and cuDNN 8.6 from. You might see CUDA 12.2 in nvidia-smi, but that should not be a problem.

Thanks! I’m going to try this.

1 Like