Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

Super Xander

macrumors 6502
Original poster
Nov 6, 2016
313
131
Denmark
This post is mostly reflecting on the rumors and potential outcome of AI performance for the upcoming iPhones. Most people do expect new AI features coming to Apple’s software at WWDC. Other rumors are also speculating that the upcoming iPhones will be some kind of an AI-phone with additional on device power for AI + a seperate feature set.

I could be a little naive, but it seems like most people are determine AI on device performance as number of TOPS. The new iPad with M4 has around 38 TOPS, and I think that a lot of people are expecting the A18 processers to be similar to the new M4 architecture since this is a new core design.

What doesn’t seems aligned is that the A17 Pro are already capable of 35 TOPS, only 3 TOPS from the M4 (and this is just a phone). Do we expect A18 to be similar to M4 and keep the 35-38 TOPS of performance making it actually similar to the current set of iPhones in terms of AI performance, or do we expect a massive improvement following what the rumors say?

I believe we are going to see similar performance as the M4 in terms of AI capability, hence similar AI performance as the current iPhones with A17. It wouldn’t make of a much consistent lineup if the AI performance on upcoming iPhones will top at 45-50 or even 60 TOPS making the upcoming Macs rumored with M4 not as capable of an AI machine compared to these small iPhones. The same with iPads.

So what will be interesting to see in the next months is how this “inconsistensy” will turn out. Will the iPhone 16 have similar AI performance as the M4 and hence the current iPhones? If the rumors of branding the new iPhones as an “AI phones” with some prioritized AI on-device features compared to other iPhones turns out to be true, then how is the current set of iPhones not just as much AI phones as the new ones?

Another thing is that the M3 never got the same AI love as the A17 Pro. A lot of people believes the chips to be “in the same family” since their design and processing node (N3B) are similar. But M3 only got 18 TOPS compared to A17 Pro’s 35 TOPS. Interesting decision by Apple since iPhone 15 Pro isn’t branded as much as an AI-phone compared to Macs as “AI computers” even though the AI performance is twice as fast on the phone.

It will be interesting to see, especially what (if any) kind of love the iPhone 15 Pro is going to get in terms of on-device AI features compared to older iPhones, the upcoming iPhone and the new iPad.
 
To round it up. I think we will get another “Siri” moment, where Siri at the time only was for iPhone 4S (and then iPad 3) even though it turned out through jailbreaks and other hacks that iPhone 4 and iPad 2 were more than capable running Siri (as Siri is more or less just a service).
 
Apple is allowing on-device AI on M1. This suggests the actual requirement for Apple is 11 TOPS (A14) with 8GB RAM.

Snapdragon X Elite is already 45 TOPS. Apple should match or come close to that, not because of marketing, but because the engineers in both companies likely project the same amount of throughput is needed for future AI apps.

Having inconsistency in NPU performance between iPhone and iPad/Mac isn't important. Apple's bread and butter remains the iPhone. That product should always get the best possible performance within the thermal envelope.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.