Theoretically, the iPhone 15 Pro and iPhone 15 Pro Max (curr. $984.89 refurbished on Amazon) are capable of delivering up to 35 TOPS of neural processing unit (NPU) performance courtesy of their A17 Pro SoC. With that being said, a theoretical integer does not actually reveal how fast an NPU could execute a given machine learning model, particularly given that there are other variables at hand like available RAM, RAM speed and software optimisations.

To that end, Apple announced at this year's World Wide Developers Conference (WWDC) that it had developed iOS 18 to take better advantage of the NPU inside theA17 Pro. For instance, it has integrated a new framework called MLTensor, which helps software like GPT2 and SSD MobileVit v2 to run up to 270% and 600% faster, respectively, in iOS 18 than in iOS 17.

So far, Apple has only publicly shared early iOS 18 beta builds. Nonetheless, the improved NPU performance is already visible in synthetic benchmarks like Geekbench ML. For example, the iPhone 15 Pro Max typically scores just over 5,500 points in Geekbench ML 0.6.0 in iOS 17.5.1. By contrast, this score jumps roughly by 2,000 points to over 7,500 points on iOS 18, a 36% improvement. Apple has also confirmed that iOS 18 will deliver better NPU performance for A16 Bionic-powered iPhones. However, its 17 TOPS NPU means that Apple Intelligence will be limited to the iPhone 15 Pro and iPhone 15 Pro Max, as well as their successors.