But its way beyond the point where its going to help LLMs. CPU RAM is already "too slow" in machines big enough for multiple NVMe SSDs.