Good for a database, maybe.
What else?
I tested out running Llama on a 512GB machine, it's rather slow and inefficient. Maybe 1-token/sec.