I have a desktop computer (with i7) that runs Ubuntu 18.04. I am using Docker to run a container that trains a model using Keras. The issue described below persists even when not using Docker.
The training process is divided into multiple steps, some of those steps utilize just one core, others run on multiple cores (
top shows 100% and 400% CPU usage respectively).
When I login locally (using monitor and keyboard) everything runs as expected and the CPU usage is the same as described above (100% and 400%).
However when I login via ssh (under the same user) the numbers are 100% and 0-18%. For whatever reason when logging in remotely the multiple cores are not utilized (not even one core is fully utilized).
What might be the issue here?