(This is related to Sage Maker Studio CPU Usage but focuses on interpreting meaning rather than modifying behaviour)
SageMaker Studio shows Kernel and Instance usage for CPU and Memory:
The kernel is just the selected Jupyter kernel and so would appear as a single process on a local machine, while the instance is the EC2 instance that they're running on.
The only documentation from Amazon appears to be in Use the SageMaker Studio Notebook Toolbar which says that it "Displays the CPU usage and memory usage. Double-click to toggle between the current kernel and the current instance" (this is outdated and relates to the old position of the information).
In the context of SageMaker Studio, does 100% CPU mean 100% of one CPU or 100% of all CPUs? (top shows multi-core as >100% but consolidated measures like Windows Task Manager's default representation show all cores as 100%)
And does 25% instance utilisation then mean that my instance is over-specced? (Intuitively, it should do because I'm not using 100% even when training a model, but I've tried smaller instances and still never maxes Instance CPU usage, only Kernel CPU usage)
I've tried using joblib to make some parallel "wheel spinning" tasks to check usage, but that just resulted in Kernel being quiet and Instance having all of the usage!