logoalt Hacker News

dist-epochtoday at 9:58 AM0 repliesview on HN

Because of LLMs we are back to memory being king (KV cache).

But there was another thing - clusters of thousands/millions of machine instead of one big iron with all the memory.