- Host must have nvida module and cuda libraries installed
- Docker container must have the nvidia devices passed in via --device flag
For instructions on how to run this, see:
I think there is an issue when setting the environment variable LIBRARY_PATH, as it should be named LD_LIBRARY_PATH. It should be set like this:
I am trying to create a similar Dockerfile for Centos 7...not too tough to translate the lines. I run into an issue on the following line
pwd/nvidia_installers && \
I get an error about /tmp not being large enough to install and the build exists.
I think you also want to apt-get install "module-init-tools":
ERROR: Unable to find the module utility
insmod; please make sure you
have the package 'module-init-tools' or 'kmod' installed. If you do
have 'module-init-tools' or 'kmod' installed, then please check that
insmod is in your PATH.
Thanks @simlay! I just fixed it.