AI features of the iPhone to be powered by the Apple M2 Ultra

Rumors have been circulating for months that the Apple iPhone and iPad will receive the biggest update in years with iOS 18 in the fall, as the new operating system version is set to introduce numerous new features based on artificial intelligence (AI). While it was previously expected that Apple's AI would run locally on the end device, the performance of the Apple A17 Pro in the iPhone 15 Pro is apparently insufficient in carrying out all the planned features.

As Bloomberg now reports, only simpler tasks will be processed directly on the ARM chip of an iPhone, such as summarizing incoming notifications. More complex tasks, such as generating an image or creating a summary of a longer document, will be processed on Apple's servers instead. Interestingly, Apple will not equip these servers with Nvidia graphics cards like most of its competitors, but with the Apple M2 Ultra, which is installed in last year's Mac Pro and Mac Studio; a new chip based on the Apple M4 is still under development.

This choice is not the most sensible from an efficiency standpoint, as the M2 Ultra achieves a machine learning performance of "only" 31.6 TOPS, while the Apple M4 achieves an AI performance of 36 TOPS, albeit with significantly lower memory bandwidth and less RAM. By comparison, the Nvidia GeForce RTX 4090 achieves an AI performance of 1,321 TOPS, more than 41 Apple M2 Ultras. The fact that Apple still uses its own chips is primarily for data protection reasons, although less critical tasks could still be outsourced to third-party servers, according to Bloomberg.