Share:

Apple M1 Impresses in Machine Learning Benchmarks

Share:

Seduced by the moniker of “fastest Mac ever” (with benchmarks to back it up), unprecedented power efficiency, and a significantly lighter body, coder and machine learning engineer Daniel Bourke found himself seriously pondering getting an Apple M1-powered MacBook Air or MacBook Pro.

Well, he bought both. Up until now, Daniel had been using a tricked out Intel-powered 16-inch MacBook Pro with 64 gigs of RAM and 2TB of storage as his daily driver. The Apple fan had paid over twice what the M1 MacBook Pro cost him for his 16-inch work computer.

Watch the Latest Apple Technology News Below

With all three MacBooks in his possession, Daniel (naturally) thought of testing them under real-world use to see whether a MacBook with an Intel chip or one with Apple Silicon would best cater to his needs, publishing the results in a blog post.

Daniel tested the three MacBooks for everything he does on his work computer — writing, browsing the web, editing videos, coding, and training machine learning models.

Typing performance across all three MacBooks is obviously the same (with the possible exception of the 16-inch MacBook Pro’s larger keyboard), and we already know the M1 Macs are decidedly faster than Intel chips at web browsing — so let’s get to the juicy bits.

The Intel-powered MacBook Pro beat both M1-powered MacBooks in video exporting, likely because of its dedicated 8GB GPU and the fact that Daniel has four times as much RAM on the 16-inch MacBook Pro than on the 13-inch M1-powered model.

The 16-inch model did, however, use up 36% of its battery during the test, whereas its 13-inch opponent took 11 minutes longer for the export but only used 9% of its battery.

Machine learning is where the M1 MacBooks absolutely shine, found Bourke. This can largely be attributed to the 16-core M1 neural engines in both models.

In a machine learning model training experiment in Apple’s CreateML, the 2020 MacBook Air ran through a hefty multi-class image classification problem in 11 minutes and 30 seconds while only using only 4% battery — the best of the bunch.

The 13-inch MacBook Pro processed the same model in 15 minutes and 30 seconds while using 7% battery. The 16-inch Intel-based MacBook Pro, however, ran the model for 43 minutes and 10 seconds before depleting the available 31% of its battery and dying before finishing.

Running a basic convolutional neural network (CNN), a transfer learning model with EfficientNetB0, and a TensorFlow benchmark all on the macOS fork of TensorFlow, the two M1-powered MacBooks posted virtually identical results and blew the Intel-powered MacBook right out of the water in everything except the TensorFlow benchmark, but couldn’t keep up with the same tests running on a GPU-powered Google Colab.

While the M1 chip may not be entirely ready for developers just yet, the processor’s machine learning capabilities are pointing to an exciting future.

Share: