More performance testing and running into traps

Wanted to better understand the performance on the LLM machine. So started putting a dashboard together.