Technically- Can Apple Intelligence work on iPhone 14 Pro hardware?

[1800AirTAG]

Besides the 33% increase in RAM fro the A16 Bionic to the A17 Pro, there's also the change from a 5nm process to a 3nm process (40% reduction in size) and the neural engine improvements ... from 17 TOPS (trillion operations per second) in the A16 Bionic to 35 TOPS in the A17 Pro.

So, we have a 33% increase RAM, a 40% reduction in process size, and greater than 100% increase in neural engine performance (which the AI features are going to lean extremely heavily on).

Those are major, directly relevant improvements. The on-device AI functionality they showed is extremely demanding and I'm certain if they enabled it on the prior phones while it might ... 'work' ... it would probably result in a laggy, jittery experience across the device.

I'm just guessing here, but given the sudden explosion of AI on the scene, and how fast it's moving, I suspect Apple was kind of backed into a corner and had to come out with something 'now'-ish. And whatever they came out with couldn't just be some token chat integration, it had to be comprehensive, robust, and amazing. I'm sure Apple would've loved to have waited another year or two to really nail things down and polish it completely ... which would've also had the side effect of having more years worth of hardware capable of running the new software.

But, they didn't have that time.

Total coincidence but 9to5 just released this which explains pretty much what you pointed out… https://9to5mac.com/2024/06/12/a17-pro-vs-a16-bionic-comparison/

 

[Klae17]

IMO they will bring a lite Version of that in the future like they did for the Stage Manager before. This is not acceptable for a device 1 year old. Audience is really important at this point...

It still does all the things it was said to do from day 1. Products are not entitled to new features and millions of people will accept it and buy new iPhones.

 

[twintin]

Do not know for sure but - I have seen some speculation that Apple managed to squeeze their LLM into 3GB RAM footprint (which is actually incredible to begin with). Seeing that you need to have RAM for OS and apps, I would not think that AI is coming to anything lower than 6GB RAM devices, at all.

I wonder what this means for iPhone 16 line of devices; will they sell a non-Pro devices with 6GB still and so AI will be a "Pro" feature? What about next SE?

There was a research paper not long ago where Apple described a method to run LLM on iPhones using flash memory instead of RAM, so I think AI may come to more iPhones later on depending how far they have come with the research.

Apple Develops Breakthrough Method for Running LLMs on iPhones

Apple GPT in your pocket? It could be a reality sooner than you think. Apple AI researchers say they have made a key breakthrough in deploying large...

www.macrumors.com

 

[aakshey]

Total coincidence but 9to5 just released this which explains pretty much what you pointed out… https://9to5mac.com/2024/06/12/a17-pro-vs-a16-bionic-comparison/

I have seen the 9to5 article. The guy is an idiot.

Going by his logic AI wouldn’t come to M1 either. But indeed it’s coming. 14 Pro has everything Apple AI needs except the RAM. Many devices lack the neural capability. 14 Pro, not so much. The only issue is RAM.

 

[TimFL1]

ITT. People blatantly underestimating how resource hungry LLMs are. It‘s not at all similar to the stage manager situation, these features wont ship for older devices and will probably run like crap on the 15 Pro / 8GB devices compared to future ones (this September, probably a supercycle).

 

[twintin]

I have seen the 9to5 article. The guy is an idiot.

Going by his logic AI wouldn’t come to M1 either. But indeed it’s coming. 14 Pro has everything Apple AI needs except the RAM. Many devices lack the neural capability. 14 Pro, not so much. The only issue is RAM.

Going by this it does mot look like RAM by se is necessarily an issue.

Apple Develops Breakthrough Method for Running LLMs on iPhones

Apple GPT in your pocket? It could be a reality sooner than you think. Apple AI researchers say they have made a key breakthrough in deploying large...

www.macrumors.com

 

[Frantisekj]

Do not know for sure but - I have seen some speculation that Apple managed to squeeze their LLM into 3GB RAM footprint (which is actually incredible to begin with). Seeing that you need to have RAM for OS and apps, I would not think that AI is coming to anything lower than 6GB RAM devices, at all.

I wonder what this means for iPhone 16 line of devices; will they sell a non-Pro devices with 6GB still and so AI will be a "Pro" feature? What about next SE?

I don't think they will release any more 6GB iPhones, or at least ones that aren't AI compatible. The iPhone lineup is the only one where AI is currently locked to the Pros.

The SE has historically matched the specs of the latest standard iPhone regarding chip and RAM with the exception of the 2nd gen. Since it will likely be staying around a few years I can see them wanting to future proof it a bit and doing that again.

I think Apple proved be blind here. They may have been working on Ai but has no clue what it will require so the have failed to make iPhone future proof by their dumb RAM policy or worse by savings and purpose.

I proved myself wrong when I was saying Apple has advantage that has many devices where they can deploy AI but it is valid only in PCs not phones. I was wondering that they could bring AI to rest of devices via their own AI servers but it would require a lot of investments. They can surprise us next year but we all doubt that I guess

 

[CrysisDeu]

I know that. What I am questioning is Apple perfectionist approach you can have everything or nothing.

Voice transcript doesn’t necessarily runs an LLM, LLM consumes a lot of memory because you need to load the large model.

Although I doubt all of Apple Intelligence functions relies on large model though

 

[CrysisDeu]

Going by this it does mot look like RAM by se is necessarily an issue.

Apple Develops Breakthrough Method for Running LLMs on iPhones

Apple GPT in your pocket? It could be a reality sooner than you think. Apple AI researchers say they have made a key breakthrough in deploying large...

www.macrumors.com

I think what this article is saying is that if this new method was not used, Apple Intelligence could not run on device, not for iPhone 15 pro and probably not even m3 pro with smaller memory

 

[CrysisDeu]

ITT. People blatantly underestimating how resource hungry LLMs are. It‘s not at all similar to the stage manager situation, these features wont ship for older devices and will probably run like crap on the 15 Pro / 8GB devices compared to future ones (this September, probably a supercycle).

I am really worried about how much background tasks they need to kill on 15 pro to run the 3B on device LLM.

They better unlock the base M4 ipad pro from 8gb to 12gb for free and bring at least 12gb to iPhone 16 pro

 

[shenfrey]

Besides the 33% increase in RAM fro the A16 Bionic to the A17 Pro, there's also the change from a 5nm process to a 3nm process (40% reduction in size) and the neural engine improvements ... from 17 TOPS (trillion operations per second) in the A16 Bionic to 35 TOPS in the A17 Pro.

So, we have a 33% increase RAM, a 40% reduction in process size, and greater than 100% increase in neural engine performance (which the AI features are going to lean extremely heavily on).

Those are major, directly relevant improvements. The on-device AI functionality they showed is extremely demanding and I'm certain if they enabled it on the prior phones while it might ... 'work' ... it would probably result in a laggy, jittery experience across the device.

I'm just guessing here, but given the sudden explosion of AI on the scene, and how fast it's moving, I suspect Apple was kind of backed into a corner and had to come out with something 'now'-ish. And whatever they came out with couldn't just be some token chat integration, it had to be comprehensive, robust, and amazing. I'm sure Apple would've loved to have waited another year or two to really nail things down and polish it completely ... which would've also had the side effect of having more years worth of hardware capable of running the new software.

But, they didn't have that time.

Good summery.

I do wonder if even the iPhone 15 Pro will struggle performance wise, as it sounds like it's probably only just fast enough.

 

[86Hawkeye]

It's not biting them in the butt, if anything it's their marketing to upgrade to an iPhone 15 Pro/16 Pro

Exactly. It’s purposeful obsolescence to sell more iPhones. It’s probably a BS excuse anyway. They’ve simply decided to lock it out of anything older than 15 Pro.

 

[HouseLannister]

Technically, sure. I think Apple Intelligence could probably run on a bunch of devices that don't support it. But it must not run it well enough to meet Apple's quality standards, and that is why they gated it where they did. They don't want to get accused of slowing down devices again, so they wouldn't roll it out if it took 30 seconds to get a response or if it caused the phone to lag horribly. They also don't want to have to hold back future development on it by having to support older hardware. An iPhone 14 Pro might have the hardware to handle the Fall rollout in 2024, but they want enough headroom to handle 2025 and 2026.

 

[Batt4Christ]

Looks like Apples stinginess on RAM has come back to bite them in the butt

Hardly. It’s well-calculated planned obsolescence. Period.