Apple A11 Bionic

What it is

The A11 Bionic is a smartphone chip that Apple announced on 12 September 2017 alongside the iPhone X^[1]. Apple presented it as the most powerful and smartest chip ever in a smartphone^[1]. The claim rested on several parts working together on one piece of silicon: a six-core CPU, an Apple-designed graphics processor, and a component Apple called the neural engine, built specifically for machine learning^[1].

A CPU with two kinds of cores

The CPU has six cores split into two groups. Two are performance cores, which Apple says are 25% faster than the A10 Fusion, and four are efficiency cores, which it says are 70% faster^[1]. Splitting cores this way is common in mobile chips: the fast cores take on demanding work when it arrives, while the smaller cores handle lighter tasks. What Apple highlighted was the second-generation performance controller, which can harness all six cores at the same time, for up to 70% greater performance on multi-threaded workloads^[1]. Multi-threaded workloads are jobs that can be broken into pieces and worked on in parallel, so having more cores available at once matters more than the speed of any single core.

Graphics

The A11 Bionic also integrates a three-core GPU designed by Apple itself, delivering up to 30% faster graphics than the previous generation^[1]. A GPU is the part of a chip that draws images and runs the kind of highly parallel arithmetic that graphics and 3D rendering depend on. Designing the GPU in-house meant Apple controlled both the CPU and the graphics side of the chip.

The neural engine

The neural engine is the part of the A11 Bionic built specifically for machine learning, augmented reality and immersive 3D games^[1]. It is a dual-core design that performs up to 600 billion operations per second for real-time processing^[1]. That figure describes throughput: how much arithmetic the block can carry out in a second, which is the relevant measure when a trained neural network has to be evaluated quickly enough that a user does not notice a delay. Apple says the engine is designed for specific machine learning algorithms and enables Face ID, Animoji and other features^[1].

Face ID as a worked example

Face ID shows what the neural engine is for. The system projects more than 30,000 invisible infrared dots, and both the infrared image and the dot pattern are pushed through neural networks to build a mathematical model of the face^[1]. All of that face processing happens on the device rather than in the cloud, and the saved data is protected by the secure enclave, a protected region of the chip set aside for sensitive information^[1].

Where it sits

The A11 Bionic is notable less for any single number than for the combination: general-purpose cores, a graphics processor Apple designed itself, and a separate block dedicated to neural-network arithmetic, all on the chip that runs the phone. Apple's framing put machine learning inside the device rather than in a data centre, and pointed to features that depend on it, from unlocking a phone with a face to animated characters that track expressions^[1]. The neural engine's specifications — dual core, up to 600 billion operations per second — are the terms in which Apple described that shift at the time^[1].