I can't see the bog-standard iPhone 16s being faster than the 15 Pros. Maybe they'll just build an "A17 regular" which is like the A17 Pro, but on a different process?
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
iPhone 16 Rumored to Feature 'Significantly' Upgraded Neural Engine for iOS 18's Generative AI Features
- Thread starter MacRumors
- Start date
- Sort by reaction score
I actually think it is quite likely that both the 16Pro and 16 will use the A18, because the cultural sensation of LLM AI probably perturbed Apple's roadmap and led to them taking extreme defensive measures.
The doubling of the capacity of the 15 Pro vs the 14 Pro hints that they already made a big leap forwards...
What are they using all these neural engine cores for now? The photo memory and face recognition hasn’t changed much in the latest gens (photo memories only for the worse in latest iOS). I don’t mean it sarcastically, I would just like to know if all these neural engine cores add anything meaningful (compared to just more general processing power). Anybody knows?
Face recognition is still horrendous on photos
Misses tons of photos and really struggles with people who are related (and of course have similar features)
I gave up on it long ago
Well the thrust of this and similar news stories is "You ain't seen nothin' yet..."
However, I expect more efficient power usage of the CPU and GPU cores even with the upgraded neural engine. That's why the iPhone 16 Plus going to a 4,003 mAh battery pack will not affect battery life per chage on the iPhone 16 Plus.
"Because of stuff"*
*Stuff: a custom neural engine capable of processing Siri's LLM entirely on device instead of sending it to the cloud. Yes, if Apple is going for privacy (they are), then on device generative AI will solve that, since none of it needs to go to the cloud.
But that requires powerful enough custom silicon. This isn't an artificially withheld feature like some that people have (rightfully) complained about. If you want a cloud based LLM, then install Google Gemini or chatGPT. They're great, but they're not private, they have a up/down delay and are susceptible to traffic/busy periods.
That said, I do believe Apple will want immediate market penetration, rather than waiting years for enough hardware turnover. They need to get in the game now, so I think they'll update cloud based Siri for devices going back several generations, including iPhones, iPads, Watches, AirPods Pro and HomePods as well as Macs. The iPhone 16 will have a custom neural engine that'll run it all on-device for privacy and offline support.
The current Pros are already really impressive vs everything else on the market... It'll be interesting to see what is next, but I'm sure Apple will make some use of the 15 Pros capabilities come September.
It's bad that iOS is including generative AI features. Those should be left out like crypto and NFTs were
Generative AI isn't just making cringy fake pictures. It's about generating on-the-fly full blown conversations – not pre-written – which is what enables the illusion of intelligence and self-awareness, like you're speaking with a person but one with instant access to all of the internet's knowledge. This isn't a fad, it's already being used by millions of people with practical applications.
Sure, I’m waiting for info on where Apple stole the data they are training the AI with. Did they do a Meta and promise they wouldn’t use our data for their own gain and then do it anyway, or did they just scrape everything publicly available like OpenAI.
Or in my case, hate everything they loved five years ago.
Apple is more likely to do what Adobe has done and license some large, reliable content sources like The NY Times, National Geographic, and Getty Images to train their models on. They don't want the liability of using unlicensed material.
On device AI? That would be the killer feature no other manufacturer has. Almost all of the heavy lifting is in the cloud. The non private, data hoarding cloud.
if Apple has to bring out new features to iPhone 6 exclusively, I hope it is that edge computing is only capable on the phone, but that cloud AI is available for the other devices. Much like how Hey Siri was only available at first when you were plugged in, unless you had the newest device in which case you could just say the command without having to be plugged in to power.
Last edited:
That would be ideal, but the Adobe case also proved that it’s not easy to compete with a model that just stole all the number one best peoples’ work.
And there really should be an iPhone 16 SE Mini 128GB for $399.
Probably make sense for the SE to be on Apple's Modem or whatever cheaper modem they find.
Don't care about the Neural Engine stuff that much, but curious to see what design improvements the p-cores get. Between A14 and A17, we only have a 10% IPC increase on iPhone, and between M3 Max and M1 Max, only 1.6%. (Or 3.1% on the M3 / M1.) Granted, these stats are a bit misleading, since Geekbench still doesn't show the clock rates separately for heterogenous cores. But nonetheless, my impression is that the p-core design hasn't changed much since the A14 (whereas the e-core design has).
Will the A18 be the year p-cores get design changes?
Will the A18 be the year p-cores get design changes?