@entron: These images were developed before the official images and allowed (with restrictions) CUDA on Docker. These are now deprecated in favour of the official images.
cuDNN licenses pre-R4 did not allow redistribution, hence I couldn't include them. However R4 does allow redistribution, so once that is no longer in "release candidate" status I will be updating all the deep learning images with cuDNN support.
seems it does not include cuDNN library? So all the kaixhin/cuda-caffe as well as kaixhin/cuda-digits does not build with cuDNN?
To address your comments:
- If you visit the Source Project Page link on the right you'll be directed to the GitHub repository.
- The PATH variable is set up using the ENV command.
- The 64-bit libraries are added using
ldconfig- can you confirm the 32-bit libraries need to be added as well (this was not mentioned in the official instructions)?
- The host driver requirement was in the description rather than the usage section, so I've now moved it to a new requirements section to make it clearer.
If there are any remaining problems please go ahead and raise an issue on GitHub.
One more thing we need to add in Usage:
Make sure host drivers use NVIDIA-Linux-x86_64-346.46.run. I have confirmed that the host driver must match with driver comes from 7.0.28 cuda package.
Also, add /etc/ld.so.conf.d/cuda.conf
It would be great if bash profile setup PATH variable to /usr/local/cuda/bin
Will you share your project to github?