Alleged Apple M4 Geekbench scores show incremental improvement in machine learning over last gen

Apple M4 Neural Engine screenshot
(Image credit: Apple)

Apple yesterday announced the latest iteration of its own silicon with the M4-powered iPad Pros. The company claims that the CPU is 50% percent faster than the M2 on the 12.9-inch iPad Pro (6th-generation), sporting two more cores (4 performance + 6 efficiency) over the last generation’s eight cores (4 performance + 4 efficiency). Although we cannot yet confirm these claims, the M4’s Neural Engine was tested today on Geekbench, with machine learning (ML) benchmarks posted under iPad 16,3. As ever with leaks and unverified benchmarks, take the news with a pinch of salt.

The results do not explicitly say iPad Pro (13-inch), but the specifications noted on the benchmark match that of the higher-end M4-powered iPad: a 10-core ARM processor with 16GB RAM. We also compared it with the 12.9-inch iPad Pro (6th generation)’s machine learning test results to see how it would stack up against, and the results show an incremental improvement over the last gen’s ML performance.

The M4 iPad Pro hit a Geekbench ML 0.6.0 score of 9234, around 22.9% better than the M2-powered iPad Pro’s 7511. We also compared the M4 iPad Pro’s ML performance with an M3 14-inch MacBook Pro, where its gains were a little more modest at 10.4% (8365 vs. 9234). However, the M3 MacBook Pro might have a slight advantage over the M4 iPad Pro owing to its larger chassis and battery.

The original M1 Apple silicon was a true generational leap over Intel’s x86 processors. However, the subsequent M2, M3, and M4 chips are just building on the M1’s architecture instead of offering massive gains. Nevertheless, the M4 focuses heavily on AI processing, with its 16-core neural engine supporting up to 38 trillion operations per second (TOPS). In comparison, the previous-gen M3 also has a 16-core neural engine. Apple's M3 was rated for 18 TOPS at FP16 precision, but the M4 is rated for 38 TOPS with INT8. That means, if equalized to INT8 precision, we're looking at a 5% improvement in TOPS for the M4 over the M3.  

However, the May 7 presentation mainly focused on hardware — the new iPad Air, iPad Pro, and the Apple Pencil Pro. We’ve only received a few snapshots of what AI features the M4 silicon will bring us, mostly on its use with Final Cut Pro for iPad 2 and Logic Pro for iPad 2. Aside from that, there wasn’t much mention of AI; we expect that to happen during WWDC 2024 instead.

Apple still has the upper hand in mobile computing, especially as Windows laptops still lag behind in terms of efficiency. But Qualcomm isn’t far behind, with the Snapdragon X Elite and X Plus chips beating AMD and Intel’s top laptop processors. Assuming Windows 11 (or Windows 12) can take proper advantage of these new Arm processors, we’re entering a new golden age of powerful yet highly efficient laptops, and Apple has to step up its game to remain in the lead.

Freelance News Writer
  • Metal Messiah.
    Apple's M3 was rated for 18 TOPS at FP16 precision, but the M4 is rated for 38 TOPS with INT4. That means, if equalized to INT8 precision, we're looking at a 5% improvement in TOPS for the M4 over the M3.

    Not entirely sure if that info is 100% correct. 38 TOPS with INT4 ? Got proof ?

    First off, Apple has NOT confirmed nor listed the exact precision info, as to whether the value is based on INT16, INT8, or even INT4 precision.

    But for ML inference, INT8 would be the most likely option/candidate, since this is what Apple quoted for the A17 chip last year.

    So the 5% improvement claim would also not be accurate.
    Reply
  • PaulAlcorn
    Metal Messiah. said:
    Not entirely sure if that info is 100% correct. 38 TOPS with INT4 ? Got proof ?

    First off, Apple has NOT confirmed nor listed the exact precision info, as to whether the value is based on INT16, INT8, or even INT4 precision.

    But for ML inference, INT8 would be the most likely option/candidate, since this is what Apple quoted for the A17 chip last year.
    Ooops, you are correct. Apple has confirmed to analysts that it is INT8. fixed the typo.

    1787861586285298111View: https://twitter.com/BenBajarin/status/1787861586285298111
    Reply
  • cknobman
    Drink the KoolAid, dont ask questions, and open your wallet.
    Reply
  • brandonjclark
    For those of us who would like to know more...

    https://www.mathworks.com/company/technical-articles/what-is-int8-quantization-and-why-is-it-popular-for-deep-neural-networks.html
    Reply
  • thisisaname
    Or just wait for a independent review of the device, with some nice benchmarks.

    Leaks are nice to speculate on but I would not trust them.
    Reply
  • TechyIT223
    Yesh should better wait for proper third party benchmarks for sure.
    Reply
  • TechyIT223
    Btw, now based on the 38 Tops INT8 metric the claimed performance should be much more than the 5% value the original article claimed ? Right?

    Or something else entirely?
    Reply
  • Metal Messiah.
    Another entry spotted :

    1788324153797615687View: https://twitter.com/VadimYuryev/status/1788324153797615687
    Reply