logoalt Hacker News

coder543yesterday at 12:36 AM0 repliesview on HN

The Spark has more compute, so it should be faster for prefill (prompt processing).

The M4 Max has double the memory bandwidth, so it should be faster for decode (token generation).