The Apple M2 for Machine Learning

How does the new M2 chip from Apple do with machine learning?


Steve Kelly

6/14/20223 min read

The recently announced M2 CPU from Apple is expected to beat the previous year's M1 chip in many ways.

For one, the new M2 has leaped bounds ahead of the M1 chip, leading to higher levels of computing power.

With machine learning applications, you can expect the M2 to blow them out of the water. The deepest neural network will have little trouble processing on the M2 processor.

TensorFlow, PyTorch, and Keras, very common machine learning frameworks will have no trouble with the new MacBook series M2 for machine learning.

We can say this with confidence as the previous M1 had no issue with running machine learning frameworks, and it is expected to be many times faster due to the increased specs on the brand new processor from Apple.

Towards Data Science has a lot of information on the previous M1 chipset for data science, and machine learning applications. I would highly recommend anyone looking for a comprehensive review of the M1 to head over there.

To summarize what they said in their blog post: "I haven’t been this impressed with a new computer since I first switched from hard drive to solid-state drive."

That was for running machine learning applications on the M1.

We can expect the M2 to be running way better than the M1 due to all of the increased specifications the M2 has over its predecessor.

For most people trying to do machine learning on the M2 chipset, there will be no issue with running the algorithms and programs. The speed will be fast, even faster than the M1. If we had to guess, no one will be able to fully utilize the processing power of the M2 chipset with their machine learning, data science, and deep learning tests.

We look forward to the release of the new MacBook series and are very excited for the M2 chip to come out. It's exciting times for Apple enthusiasts, and newcomers alike.

At this time, the new 2022 MacBooks running the new M2 have been released. They were released June 24th 2022. The M2 Pro and Max were released January 17th 2023. The M2 Ultra was released June 13th 2023.

If you're not interested in going with an Apple product for machine learning, Laptop Study has provided a list of alternatives for computer science focused laptops which includes machine learning.

Update 8/31/2023: With the latest models, the M2 Pro, the M2 Max, and the M2 Ultra being released we figured it would be time to give a bit of an update. While the world awaits note of the M3 chipset being announced, the latest upgrades should provide plenty of computing power for machine learning applications.

The M2 Pro has

  • Up to a 12-core CPU

  • Up to a 19-core GPU

  • Up to 32GB of unified memory

  • 200GB/s memory bandwidth.

    Just looking at those specifications seems like insane computing power to me. Machine learning applications should have no trouble running on the M2 Pro chipset. Wait a minute..., we haven't even gotten to the M2 Max or the M2 Ultra yet.

The M2 Max

  • Has a 12-core CPU straight out of the gate

  • Up to 38-core GPU

  • Up to 96GB of unified memory

  • 400GB/s memory bandwidth.

    Whew. That's mind-boggling power for machine learning capabilities. There's just no getting around it, Apple is providing a ton of computability for its price point.

The M2 Pro and the M2 Max are available on the MacBook Pro 14" and 16" models. Well, what about the latest and greatest M2 Ultra?

The M2 Ultra is available on the Mac Studio and the Mac Pro. While these are not laptops, they are powerful computers that will surely handle any machine learning tasks you throw at it with ease. Alright, let's get down to the specifications for this powerful chipset. The Ultra has a feature-set tuned for machine learning, according to Apple: "M2 Ultra features a 32-core Neural Engine, delivering 31.6 trillion operations per second, which is 40 percent faster performance than M1 Ultra."

The specifications for the M2 Ultra are:

  • 24-core CPU

  • Up to 76-core GPU

  • 192GB of unified memory

  • 800GB/s memory bandwidth

    I believe with this level of computing power you could run large language models locally on the M2 Ultra. TechRadar reports that Apple is rivaling its rivals such as AMD in computing power. Considering all of the above, you can easily get a ready to go machine able to tackle machine learning tasks and artificial intelligence tasks with Apple products. We eagerly await Apple to announce their own version of a large language model (Siri may be getting an upgrade soon).