How fast are LLM inference engines anyway?

This YouTube talk is overlaid with portable W3C Web Annotations connecting it to its canonical devreal.ai graph nodes.