Old RTX 3090 enough to serve thousands of LLM usersA single RTX 3090 is sufficient for serving smaller language models to thousands of users, challenging the notion of needing enterprise GPUs.