Hi,
I’m working with a model built on Tensorflow 1.13.1 which requires I install cuda10.0 and cudnn7.5. After some trouble getting this installed correctly (working through Installation Guide Linux :: CUDA Toolkit Documentation), I uninstalled all cuda/cudnn from my machine, then tried the following.
Below is the RUN command from Nvidia’s cuda10.0/cudnn7 dockerfile[1] (except I added ‘sudo’ a few places where needed) rolled back to before this commit[2] since I want cudnn7.5 not cudnn7.6:
CUDNN_VERSION="7.5.1.10"
sudo apt-get update && sudo apt-get install -y --no-install-recommends \
libcudnn7=$CUDNN_VERSION-1+cuda10.0 && \
apt-mark hold libcudnn7 && \
rm -rf /var/lib/apt/lists/*
^ But this returns this error: E: Version '7.5.1.10-1+cuda10.0' for 'libcudnn7' was not found
Does this mean Nvidia isn’t maintaining a cudnn7.5/cuda10.0 image available for installing via dockerfiles? I’ve been struggling getting all my nvidia/cuda/cudnn environment set up correctly, and was hoping borrowing from the dockerfile would streamline things. I assume this means there’s no point trying to use docker/nvidia-docker for my cuda10.0/cudnn7.5 needs either?
Thanks,
Max
[1] https://gitlab.com/nvidia/cuda/blob/ubuntu18.04/10.0/runtime/cudnn7/Dockerfile
[2] Update cuDNN to 7.6.0.64 (215f0518) · Commits · nvidia / container-images / cuda · GitLab