We Already Know How Well Apple's M4 Chip Performs
While new iPads are always exciting, it was the iPad Pros' M4 chip that stole the show this week. M3 just came out in October, and Apple decides to launch its successor now—in an iPad, of all devices. Still, the chip is here, and is likely what will power Apple's big AI plans this year.
Apple touted M4's 3-nanometer architecture, its four performance cores and upgraded six efficiency cores, and claimed it can deliver up to 1.5 times faster CPU performance over M2. (The company does prefer to compare its chips to two generations prior, rather than one.) Of course, the iPad isn't out yet, so reviewers and users can't test these claims. While we could just take Apple's word for it, we could also just look to Geekbench.
The M4's stats starting appearing on Geekbench, the popular benchmarking site, on Wednesday. It's not clear who is testing the M4 through Geekbench, but it offers an interesting window into the chip's performance. At the time of this article, single-core performance ranges from 3,595 all the way to 3,824, while multi-core performance ranges from 13,910 to a whopping 14,748. (Single-core refers to testing one of the chip's cores, while multi-core refers to testing multiple cores at once. For context, the M4 has 10 total CPU cores.)
The iPad Pros being tested are running iOS 17.5, not yet released, with a base clock speed of 4.4 GHz. They're equipped with 16GB of RAM (although it comes through as 15.05GB on Geekbench).
These scores are impressive, especially considering they're running on iPads (and the thinnest iPads ever made, at that). Based on these scores, M4 modestly outperforms M3. If you look at Geekbench's scores for the M3 MacBook Pro (not M3 Pro or M3 Max, mind you), you see a single-core score of 3,079 and multi-core score of 11,524. If you look at the 6th gen iPad Pro running an M2 chip, the difference is more stark: Geekbench shows this tablet scores 2,613 in single-core, and 10,082 in multi-core. That's a near one-and-a-half times jump in performance, which tracks close to Apple's claims.
I'm interested to see how M4 performs once Apple puts it into a Mac. It's possible Apple is underclocking M4 on iPad since the tablets are so thin and light, they wouldn't be able to efficiency move the heat generated by pushing the chip. Theoretically, even a fanless MacBook Air would have a better chassis for heat distribution, so M4 should perform even better. Put it in something like the MacBook Pro with a dedicated fan, and the results should only improve, at least once you really start taxing it.
But regardless, my main focus now is on the M4's NPU (neural processing unit). This part of the chip is responsible for processing AI, which Apple is likely to focus on big time following WWDC. The company says M4's Neural Engine is capable of 38 trillion operations, and is 60 times faster than the company's first NPU they built for the A11 Bionic chip. (Again, Apple is comparing stats to a chip they released in 2017 for the iPhone 8, 8 Plus, and X.) While I'm sure the M4 is going to be the best way to experience all the AI features Apple is cooking up, I'm skeptical it's going to be a huge leap over M3, M2, or even M1. Maybe there will be features exclusive to M4, but my guess is Apple's entire lineup of silicon chips will be capable enough to run most of the new AI features in iOS 18 and macOS 15.