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).