If I write a regular (non-AI) program and try to run it inside a Nvidia GPU Docker container, will it force my program to use the GPU device on my server, and thus accelerate my program?
Note: my program wouldn’t include any CUDA code or pragmas.
If your application has no CUDA acceleration outside of a docker container with nvidia runtime enabled, then your application will also have no CUDA acceleration inside of a docker container with nvidia runtime enabled.
Specifying --runtime=nvidia (after setting up your docker daemon.json file in /etc/docker, I believe) only makes it so that your application can use the gpu acceleration inside of a container; it doesn’t actually default to using it.