Apple Intelligence "can't" run on an iPhone 15 (with an A16 chip) but can on an M1 chip with a much weaker NPU. The only explanation is money.

[izzy0242mr]

Apple has made the artificial decision to limit (at least for now) their AI features to devices with certain hardware—not because it's needed to run things (since they admitted in the keynote that many AI features will do cloud processing), but because they can.

I'm sure the on-device AI features will run more smoothly on chips with larger Neural Engines, which we know the AI features primarily use. But just compare the Trillions of Operations Per Second (TOPS) figure of the M1 Pro chip (11 TOPS) vs the more powerful A16 chip in the iPhone 15 (17 TOPS) and you'll realize exactly how artificial this limitation is.

To be clear, the iPhone 15 (base) has a processor with a Neural Engine/NPU (chip designed to run AI capabilities) that is 150% more powerful than a Mac with the M1 Pro chip, and yet the Mac will be able to run Apple Intelligence features while the newer, more powerful iPhone 15 will not.

EDIT: As @Ansath pointed out, the common factor between all AI-supporting devices is 8 GB (or more) of RAM. However, that doesn't change the fact that Apple said AI processing will often happen off-device in many cases…which means there's no good reason why older devices with almost as much RAM and more powerful NPUs shouldn't be able to use at least some AI features.

 

[contacos]

They are making the Non-Pro iPhone even less attractive to users. Who knows what may not be available on the regular iPhone 16 with iOS 19

 

[izzy0242mr]

They are making the Non-Pro iPhone even less attractive to users. Who knows what may not be available on the regular iPhone 16 with iOS 19

In fairness, they're clearly branding Apple Intelligence as a beta feature. They could potentially be trying to test out the products on their 15 Pro users, get more data on how to improve it, and then release it to the "masses" in iOS 19 next year. Only developers and nerds will be installing iOS 18 Beta right now; the normal iPhone 15 Pro users will get it in the Fall of 2024. Summer 2025 iOS 19 will come out, and they'll have had a good half year of time to improve AI for iOS 19. I'm probably being too optimistic.

Another clue that it might not come to the iPhone 16 is that the iPhone 15 Pro's chip is the "A17 Pro," not just the "A17." That implies there'll be an A17 "normal" chip for the iPhone 16 base model.

But once again, there's nothing physically preventing Apple from allowing Apple Intelligence on even the iPhone 12! Because that has the A14 chip which has (like the M1 and M1 Pro chips) an 11 TOPS NPU/Neural Engine. The iPhone 12 is just as capable of running Apple Intelligence as an M1 Pro Mac.

 

[Ansath]

Apple has made the artificial decision to limit (at least for now) their AI features to devices with certain hardware—not because it's needed to run things (since they admitted in the keynote that many AI features will do cloud processing), but because they can.

I'm sure the on-device AI features will run more smoothly on chips with larger Neural Engines, which we know the AI features primarily use. But just compare the Trillions of Operations Per Second (TOPS) figure of the M1 Pro chip (11 TOPS) vs the more powerful A16 chip in the iPhone 15 (17 TOPS) and you'll realize exactly how artificial this limitation is.

To be clear, the iPhone 15 (base) has a processor with a Neural Engine/NPU (chip designed to run AI capabilities) that is 150% more powerful than a Mac with the M1 Pro chip, and yet the Mac will be able to run Apple Intelligence features while the newer, more powerful iPhone 15 will not.

View attachment 2387122

According to Ming-Chi Kuo, RAM is a factor. the On Device LLM needs 8GB RAM to run. The 15 Pro's have 8Gb, the non-Pro has 6GB.

 

[Ansath]

The iPhone 12 is just as capable of running Apple Intelligence as an M1 Pro Mac.

iPhone 12 had 4GB RAM. M1 Pro Mac, has much more RAM......

It's not just about the M1 chip.

 

[Broken Hope]

They need at least 8GB of RAM to store Apple's AI model locally.

 

[izzy0242mr]

According to Ming-Chi Kuo, RAM is a factor. the On Device LLM needs 8GB RAM to run. The 15 Pro's have 8Gb, the non-Pro has 6GB.

iPhone 12 had 4GB RAM. M1 Pro Mac, has much more RAM......

It's not just about the M1 chip.

The on-device AI features definitely need more RAM, and a stronger NPU makes perfect sense for that. No dispute there.

But that's also not the point. As I noted in the OP, Apple said many AI features would be processed off-device, which means that it doesn't make sense to limit ALL of Apple Intelligence to just the iPhone 15 Pro. If they said "we're limiting SOME features of Apple Intelligence to the iPhone 15 Pro," that would make sense, but currently ALL of Apple Intelligence is limited to those devices.

 

[NT1440]

Hold on. Why are you assuming everything this needs is simply the Neural Engine?

I bet if you watch the SOTU you’ll get a bit more insight 🤷‍♂️

But hey, go off on that rant.

 

[Broken Hope]

If it was all about money why would they even support M1? Those devices came out in 2020 as I'm typing this on a M1 Mac mini.

 

[izzy0242mr]

Hold on. Why are you assuming everything this needs is simply the Neural Engine?

I bet if you watch the SOTU you’ll get a bit more insight 🤷‍♂️

But hey, go off on that rant.

I probably distracted everyone with the NPU mention. As someone else pointed out, the common factor between the 15 Pro and the M1+ Macs is 8+ GB of RAM.

However, that doesn't change the fact that Apple said much of the AI processing happens off-device…which means on-device compute power isn't always even needed.

If it was all about money why would they even support M1? Those devices came out in 2020 as I'm typing this on a M1 Mac mini.

Developers.

iPhone-driven sales (hardware + software) is the largest revenue source for Apple. They're not making a huge chunk of their profits from Mac sales. They want devs to use AI and figure out how to add AI features to their apps (they alluded to this in the keynote). Getting more AI features by developers who use Macs to make iOS apps makes more monetary sense to push AI features to the iPhone faster.

Meanwhile, pushing more people to get newer iPhones pushes that market segment a lot more. The iPhone is the central device of Apple for right now.

 

[NT1440]

I probably distracted everyone with the NPU mention. As someone else pointed out, the common factor between the 15 Pro and the M1+ Macs is 8+ GB of RAM.

However, that doesn't change the fact that Apple said much of the AI processing happens off-device…which means on-device compute power isn't always even needed.

Developers.

iPhone-driven sales (hardware + software) is the largest revenue source for Apple. They're not making a huge chunk of their profits from Mac sales. They want devs to use AI and figure out how to add AI features to their apps (they alluded to this in the keynote). Getting more AI features by developers who use Macs to make iOS apps makes more monetary sense to push AI features to the iPhone faster.

Meanwhile, pushing more people to get newer iPhones pushes that market segment a lot more. The iPhone is the central device of Apple for right now.

Okay, maybe rewatch it read the transcript of the event. MOST of the process is on-device.

Also, I stated from day one of M1 that Apple clearly designed it to be the baseline for upcoming features. So we can put to bed this nonsense that Apples work on “Apple Intelligence” hasn’t been 5+ years in the making. They’re not relying on Devs to make this useful, they’ve brought tools to Devs to incorporate the work Apple has already done.

 

[Ansath]

The on-device AI features definitely need more RAM, and a stronger NPU makes perfect sense for that. No dispute there.

But that's also not the point. As I noted in the OP, Apple said many AI features would be processed off-device, which means that it doesn't make sense to limit ALL of Apple Intelligence to just the iPhone 15 Pro. If they said "we're limiting SOME features of Apple Intelligence to the iPhone 15 Pro," that would make sense, but currently ALL of Apple Intelligence is limited to those devices.

Simple - all limited to those devices, so as to give the best experience.

Fragmenting it so that certain older devices get half the experience, would be terrible. You'd get people complaining that they're not getting the same experience as others they see on Telly, of on their friends phone, etc & most people won't get that they're getting a limited experience cos their phone is older. They'll just complain and it'll be worse press for Apple, than how they are doing it.


Plus, if you saw what was said, a lot of the on device stuff, works out what it then needs to use the cloud for - the older devices likely don't have the power to do that effort.

It's like your purposefully being obtuse/ignoring/misrepresenting what was said, so you can kick off. Which I won't waste anymore time on with you.

 

[Ww6joey]

For all we know these neural engine/npu is just marketing BS. Need the brute power of good old ram to make it run 😂

But you’re right. It’s mostly likely just about money.


/hidingmyangerwitha6montholdiphone15 😭

 

[LogicalApex]

In fairness, they're clearly branding Apple Intelligence as a beta feature. They could potentially be trying to test out the products on their 15 Pro users, get more data on how to improve it, and then release it to the "masses" in iOS 19 next year. Only developers and nerds will be installing iOS 18 Beta right now; the normal iPhone 15 Pro users will get it in the Fall of 2024. Summer 2025 iOS 19 will come out, and they'll have had a good half year of time to improve AI for iOS 19. I'm probably being too optimistic.

Another clue that it might not come to the iPhone 16 is that the iPhone 15 Pro's chip is the "A17 Pro," not just the "A17." That implies there'll be an A17 "normal" chip for the iPhone 16 base model.

But once again, there's nothing physically preventing Apple from allowing Apple Intelligence on even the iPhone 12! Because that has the A14 chip which has (like the M1 and M1 Pro chips) an 11 TOPS NPU/Neural Engine. The iPhone 12 is just as capable of running Apple Intelligence as an M1 Pro Mac.

That’s the current unknown.

It could end up being a feature line Apple uses to further segment the lineup. Where the “Pro” features like “AI” are Pro model exclusive for an extended duration. Like Dynamic Island was or the Action Button.

Today’s apple that doesn’t surprise me.

They’ll also probably announce new iPhone 16 exclusive features for it in September too. As the cycle always goes to drive upgrades.

 

[LogicalApex]

Fragmenting it so that certain older devices get half the experience, would be terrible. You'd get people complaining that they're not getting the same experience as others they see on Telly, of on their friends phone, etc & most people won't get that they're getting a limited experience cos their phone is older. They'll just complain and it'll be worse press for Apple, than how they are doing it.

Apple has been doing artificial segmentation for years. So it is fair for people to question the validity of their claims.

For instance the iPhone 15 Pro has additional camera features than the iPhone 14 Pro even though the hardware is identical. Apple has done this frequently and they make the claim that the NPU is needed for the new features. Like 35mm zoom levels 🤣.

Or the M2 being needed for Apple Pencil Hover.

 

[joey561]

The on-device AI features definitely need more RAM, and a stronger NPU makes perfect sense for that. No dispute there.

But that's also not the point. As I noted in the OP, Apple said many AI features would be processed off-device, which means that it doesn't make sense to limit ALL of Apple Intelligence to just the iPhone 15 Pro. If they said "we're limiting SOME features of Apple Intelligence to the iPhone 15 Pro," that would make sense, but currently ALL of Apple Intelligence is limited to those devices.

OP is completely right. Typical Apple feature rationing. I have a 15 Pro so will be able to use these features (whether or not I choose to is another matter) - but it’s almost certain that there is no reason this should not at least also be available for the 14 pro/standard 15.

 

[Ansath]

Apple has been doing artificial segmentation for years. So it is fair for people to question the validity of their claims.

For instance the iPhone 15 Pro has additional camera features than the iPhone 14 Pro even though the hardware is identical. Apple has done this frequently and they make the claim that the NPU is needed for the new features. Like 35mm zoom levels 🤣.

Or the M2 being needed for Apple Pencil Hover.

The camera features you’re talking about are not ones the general consumer see/care about. So, not really the greatest comparison.

These AI features are extremely more prominent.

 

[Paddle1]

It could end up being a feature line Apple uses to further segment the lineup. Where the “Pro” features like “AI” are Pro model exclusive for an extended duration. Like Dynamic Island was or the Action Button.

The fact it works on the "cheaper" M1 devices like the MacBook Air and Mac mini but not the latest iPhone is evidence against it being an arbitrary limitation. 8GB of RAM is the common denominator.

 

[brinkeguthrie]

"As you can see, all the same iPhones running iOS 17 can be updated to iOS 18. However, there’s a catch. According to Apple, only the iPhone 15 Pro and iPhone 15 Pro Max are compatible with the new AI, or Apple Intelligence, features. This is because these features require the A17 Pro or M1 chip or later."


I have an 11 Pro Max. Does this mean Siri won't work?

 

[Ansath]

"As you can see, all the same iPhones running iOS 17 can be updated to iOS 18. However, there’s a catch. According to Apple, only the iPhone 15 Pro and iPhone 15 Pro Max are compatible with the new AI, or Apple Intelligence, features. This is because these features require the A17 Pro or M1 chip or later."


I have an 11 Pro Max. Does this mean Siri won't work?

You’ll have Siri running as it does now. Just won’t have any of the Apple Intelligenfe features the talked about.

 

[kwandrews]

They need at least 8GB of RAM to store Apple's AI model locally.

A good reason to go with more RAM in newer Macs and other devices when there is a choice. So the iPhone Pro 8G RAM, iPad Pro M / iPad Air M both have 8G RAM (or higher on certain Pro models). This all seems reasonable. I think we'll start seeing Apple devices in the coming years with more base RAM to expand the pool of users....example, I suspect that even the new iPhone SE and iPhone 16 will have a min of 8GB and we'll start seeing the new MacBook Pros starting at 24GB in the future and the Air's at the current Pro min of 18GB with the M4 chips or M5's coming.

 

[Ansath]

A good reason to go with more RAM in newer Macs and other devices when there is a choice. So the iPhone Pro 8G RAM, iPad Pro M / iPad Air M both have 8G RAM (or higher on certain Pro models). This all seems reasonable. I think we'll start seeing Apple devices in the coming years with more base RAM to expand the pool of users....example, I suspect that even the new iPhone SE and iPhone 16 will have a min of 8GB and we'll start seeing the new MacBook Pros starting at 24GB in the future and the Air's at the current Pro min of 18GB with the M4 chips or M5's coming.

Yea, exactly. I was surprised the non-pros only had 6GB RAM last year, was poor planning/design by Apple.

 

[ghostface147]

They need at least 8GB of RAM to store Apple's AI model locally.

Citation please.

 

[Ansath]

Citation please.

Here’s one source for you….

投資人對萬眾矚目的WWDC 2024最常提出的問題與我的想法 / Most Frequently Asked Questions from Investors and My Thoughts…

投資人非常關注Apple將在WWDC 2024發布的AI服務和功能。以下是過去數月中投資人們最常詢問我的問題：

medium.com

If you research about LLM, ideally they need 16GB to run locally, but can make it work with 8GB as a minimum. The non-pro 15s have 6GB.

 

[Dovahkiing]

Citation please.

This is also pretty common knowledge in AI field. The LLM is stored as a model - you can go browse ollama and look at the sizes of other publicly available models that you can run on your computer locally. The smallest practically useful models are in the neighborhood of 4GB or so. The whole model gets loaded into RAM when it runs. Even if Apple developed a super efficient LLM model, it likely takes up at least 1.5-2GB or so as a reasonable guess. Factor in the OS and everything else your phone is doing and you can start to see 6GB is not enough.