nvidia/cuda:11.8.0-cudnn8-runtime-ubuntu22.04 missing /usr/local/cuda-11.8/targets/x86_64-linux/lib/libnvrtc.so
nvidia/cuda:11.8.0-cudnn8-runtime-ubuntu22.04 is missing the symbolic link /usr/local/cuda-11.8/targets/x86_64-linux/lib/libnvrtc.so to /usr/local/cuda-11.8/targets/x86_64-linux/lib/libnvrtc.so.11.2
When we run tensorflow in the docker image we get: Could not load library libcudnn_cnn_infer.so.8. Error: libnvrtc.so: cannot open shared object file: No such file or directory
If we switch to nvidia/cuda:11.8.0-cudnn8-devel-ubuntu22.04 tensorflow run as normal and we can see /usr/local/cuda-11.8/targets/x86_64-linux/lib/libnvrtc.so exists.
Also if we add the symbolic link to our docker image:
cd /usr/local/cuda/targets/x86_64-linux/lib &&
ln -sv libnvrtc.so.11.2 libnvrtc.so
tensorflow will work properly.
We been using nvidia/cuda:11.8.0-cudnn8-runtime-ubuntu22.04 with tensorflow with no issues until the update to the runtiume image on Feb 2, 2023