Apple ships the M1 with a dedicated neural engine

Apple announced the M1, the first chip it designed specifically for the Mac and the start of a two-year transition away from Intel processors. Built on a 5-nanometre process with 16 billion transistors, the M1 paired eight CPU cores (four high-performance, four high-efficiency) and up to eight GPU cores with a 16-core Neural Engine that Apple rated at up to 11 trillion operations per second, and a unified memory architecture letting the CPU, GPU and Neural Engine share data without copying it between separate pools.

Apple framed the Neural Engine’s speed-up in terms of everyday tasks — video analysis, voice recognition, and other on-device machine-learning features baked into macOS and its apps — rather than as infrastructure for training or running large models, since the M1 shipped only in consumer laptops and a Mac mini, not a workstation or server part. The chip powered the MacBook Air, 13-inch MacBook Pro and Mac mini from launch.

The M1 mattered to the AI story less for the model class it enabled than for what it normalised: dedicated neural-network acceleration became standard silicon in a mainstream consumer computer rather than a specialised add-on, at a point when the industry’s headline AI news was still overwhelmingly about ever-larger models running in data centres. That divergence — capability concentrating in the cloud even as inference hardware quietly became commonplace on the device in people’s hands — set up the on-device and local-inference arguments that intensified once open-weight models small enough to run on a laptop began appearing from 2022 onward.