How many branches can your CPU predict?

A simple loop lit up the internet: in Daniel Lemire’s blog test, AMD’s newest chip “memorized” 30,000 if‑statements, Apple’s did 10,000, and Intel’s only 5,000. Translation for non‑nerds: modern chips guess which way your code will go to run faster. When fed the same “random” pattern over and over, AMD guessed best. Cue the drama.

Commenters roasted Intel like it’s 2012 again, joking that AMD is a mind reader, Apple’s the middle child, and Intel’s got goldfish memory. But the pushback was instant: one engineer argued that making a bigger “guesser” costs chip space and speed elsewhere — maybe Intel spent those resources on other features. Another crowd questioned the test itself: if you use random values, aren’t you measuring memorization, not real‑world smarts? One dev admitted they had to switch to “branchless” code because Intel stumbled under load — a very practical “ouch.” Others scratched their heads: how can a single if‑statement be learned thousands of different ways?

So is AMD the psychic hero or just acing a memory quiz? The thread split cleanly: performance bragging rights vs. “this isn’t the real world.” Either way, everyone agreed on one thing — your code’s shape can make or break your CPU’s magic.