Apple is introducing a new central feature of iOS 18 in the form of Apple Intelligence (“AI” for short, no doubt a clever marketing ploy to appropriate this abbreviation for itself in this way now), which runs to a large extent on the devices – and of course that needs power.
So much so that only the current iPhone 15 Pro and 15 Pro Max are the only models capable of running AI. So clearly, if you want this, you either already own one of these current-generation Pro devices with the A17 Pro chip and its 16-core Neural Engine, or you’ll be buying a new iPhone 16 that meets these requirements by autumn at the latest.
But is this the only reason that Apple Intelligence won’t run on the regular iPhone 15 and iPhone 15 Plus?
Not a sales gimmick
However, Apple does not want to be accused of such a sales gimmick but explains in more detail in an interview why more power is actually required for a satisfactory use of the new AI system.
In principle, AI would also run on older devices, but not in the desired way, which is why Apple is preventing this.
This is explained by top managers from the company in “The Talk Show” recorded at WWDC 2024. John Gruber from Daring Fireball discussed Apple Intelligence and other WWDC announcements in front of a live audience with Apple’s Head of Software Development Craig Federighi, Head of Marketing Greg Joswiak and Head of AI/Machine Learning John Giannandrea.
Of course, this takes over two hours, so it’s good that Macrumors has highlighted some of the most important passages on Apple Intelligence. This is what John Gianandrea says:
So when you run these models at runtime (“run times”), it’s called inference, and inference of large language models is incredibly computationally intensive. So it’s a combination of the bandwidth of the device, the size of the Apple Neural Engine and the power of the device to run these models fast enough to be useful. Theoretically, you could run these models on a very old device, but it would be so slow that it wouldn’t be useful.
John Giannandrea, Apple
Possible on “old” devices, but very slow
Presenter John Gruber immediately interjects: “So it’s not about selling new iPhones?” Head of Marketing Greg Joswiak jumps in: “No, not at all. Otherwise, we would have been smart enough to just use our latest iPads and Macs, wouldn’t we?”
An M1 chip is sufficient for iPads and Macs, which several generations now have at least one built-in chip: The iPad Pro from model year 2021, the iPad Air came with the M1 in 2022. The iPad Pro is now at M4 and the iPad Air at M2. All new Macs since November 2020 have an M chip installed.
Only the iPhone requires the latest Pro generation or better as soon as Apple Intelligence is available to the general public. This is at least the current status, and it also seems understandable so that you don’t literally fall asleep when waiting for Apple Intelligence to process inputs.
Apple always endeavours to integrate older devices
Apple’s head of software development Craig Federighi (see his funniest moments from WWDC over the years) adds that the company’s first step with any new feature is to figure out how to bring it back to older devices as much as possible. But when it comes to Apple Intelligence, the hardware is the deciding factor: “It’s a pretty extraordinary thing to run models with this power on an iPhone,” he added.
According to Apple, the processing speed of the iPhone 15 Pro with the A17 Pro chip and the 16-core Neural Engine is up to two times faster than the A16 chip in the iPhone 15 and iPhone 15 Plus. This means that almost 35 trillion operations can be carried out per second.
Federighi also hinted that memory is another aspect of the system which the new AI functions require. It is therefore probably no coincidence that all devices compatible with Apple Intelligence have at least 8GB of RAM. The analyst Ming-Chi Kuo saw this as the real reason last week, Apple’s statements suggest a more complex overall situation.
Related reading:
This article originally appeared on our sister publication Macwelt and was translated and adapted from German.