Kinda hard to charge subscription fees with on device AI, but they figured it out."Introducing our magical, incredible iCloud AI Pro Max... starting at only $69/month.
We think you'll love it!"
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.
Apple to Power AI Features With M2 Ultra Servers
- Thread starter MacRumors
- Start date
- Sort by reaction score
So, all these “on-device processing” rumors were just rumors I guess. I mean, I would have actually liked if iPhone could autonomously think out-of-the-box without requiring any Internet connection, at least for most basic stuff or if I could download certain “memory chunks” that I would need all the time.
Imagine asking Siri “how to make tasty sandwich” during power outage and it does not say “oh I am so sorry there is no Internet”. I mean it would have been cool. Not really much AI but would need lots of data anyway.
And yeah put some strain on those useless machine learning chips: I would have said “you know Siri, I have a better sandwich recipe, what do you think ‘bout it?”. Why? Machine gotta learn something!😀
Imagine asking Siri “how to make tasty sandwich” during power outage and it does not say “oh I am so sorry there is no Internet”. I mean it would have been cool. Not really much AI but would need lots of data anyway.
And yeah put some strain on those useless machine learning chips: I would have said “you know Siri, I have a better sandwich recipe, what do you think ‘bout it?”. Why? Machine gotta learn something!😀
I'm down with this. Sounds like a good idea. Wouldn't be surprised if they add it to the Apple One Subscription though
3090/4090 are gaming GPU's, designed for gaming not AI tasks unlike Hopper. it's like buying GTX 1660Ti and then complain that it can't run ray tracing, because it's simply not made for such task.
You were talking about your M1 Max not M2 Ultra, that's where 400GB/s came from.
I wasn't talking about Nvidia's "gaming" GPU's in the first place, not sure how you got that impression in a article about AI Features.
Hopefully this rumor is just one of those typical random BS that Gurman throws but we're almost halfway through 2024 and Siri can barely set a timer without saying "Here’s what I found on the web" .
HobeSoundDarryl
macrumors G5
"Only Apple is able to integrate AI on device and iCloud subscription upcharges. We think you'll love it!"
Last edited:
You are all over the place lol. Were you comparing M1 Max with H100 or H200? I bought up ultra because of the server conversation. 4090 is a fair comparison for a prosumer workstation, with Max Max or even an Ultra. Either ways, Siri isn’t Apple AI/ML, I couldn’t care less if they can Siri tomorrow. AI at Apple is lot more than Siri, in fact you don’t really hear much about Siri in Apple AI literature/github or tech documentation.
Apple does not have any chips for servers. Neither M2 Ultra nor M4 have the I/O and RAM memory scaled properly for the servers. Another problem is that nobody develops any AI for Apple chips. It's mostly NVIDIA. I would understand if Apple were developing the servers to deploy in the cloud for third party software developers to develop AI solutions for Apple hardware but generic AI on Apple chips? That sounds... strange.
You must be new here…
They couldn’t even do that for a pencil. No way they do it for A1. Tim says buy new hardware!
But WHO is making the software?
Is it Apple AI?
Or will they partner with a 3rd party to run their software on Apple hardware?
I imagine some work can be done on-device in the background, like summarizing messages, emails and the like. Stuff like Siri, where there's a user waiting for a response, will need to go to a server to be processed faster.
Right. Generative AI is just a tool for making whatever it is you're making. Like a hammer for nails. If you don't know what you're doing in either instance you stand to harm yourself.
You are still talking about a gaming GPU and comparing it to your basement workstation, guess who is all over the place.
Apple is trying to compete in AI with H200's and soon B200's with their outdated chips (at least according to this article).
Siri is just a living corpse, ofc we don't hear about it.
What is they can attach many M2 Ultra together. Maybe they can now. How much is a h200 compare to one ultra.
Same chip architecture... both Apple and Nvidia chips use ARM base chips designs, apple decided to focus in efficiency and thermal management, so apple is more than capable to design AI chips
Probably worth remembering too that Apple has promised to be carbon neutral and rolling out solar power for their operations. Nvidia 4090 or whatever may have more power in certain situations, but if you are paying for all of that in green energy then Apples M chips start looking more attractive. Also helps Apple insulate from the AI gold rush that is coming and will drive other component prices sky high in price.
They could do the same with GPU.
The right question is, how many M2 Ultra does it take to reach b200 in AI performance ?
The TOPS benchmark metric isn't particularly useful because there's no standardization as to what constitutes an "operation per second."
The M4 does 38 TOPS using INT8 whereas the M3 does 18 TOPS using INT16, meaning it's analyzing double the data. Cut down to raw firepower the M4 only has about a 5% AI/ML advantage over the M3 when TOPS are equalized.
I feel like Apple doing their own thing is better. Everyone is using Nvidia. Some competition is needed as they have. Bit of a stranglehold on the market. Hence their share price.
Google is also using their own hardware for Gemini's LLM as far as I know, but they used TPU's, custom-designed for machine learning, not Chromebook hardware.
True but nvidia is currently charging a crazy premium.
At this point this must be just trolling. AI from large models will be a component of every job and every creative profession in a few years. It’s like calling statistics a toy and fad.
It seems crazy to use M2 chips to power AI when the M3 and M4, by Apple's own admission, come with a hugely upgraded Neural Engine. What on earth are they thinking?! Maybe there's just a huge surplus of unsold M2 Ultra chips that they need to do something with...