Macs to Get AI-Focused M4 Chips Starting in Late 2024

[FlyingTexan]

So you believe Mac service centers are currently being run using arrays of Mac Pro and Mac Studios with ultra chips? I agree that service centers are a critical profit center for Apple, I am just dubious that anyone runs a service center using banks of machines with MxUltra chips (although banks of Mac mini is something I’ve seen being used, I’m not sure what Apple uses)

What? No. Not at all. I'm saying the SERVICES are what makes the money. IE You pay $2.99 or $9.99/month for storage. Your photos and videos, your search data, etc. The app store and getting 40% of a sale. all the things they wouldn't have if you didn't have the device in your hand. The warranties on everything.

 

[FlyingTexan]

Here is what you and most here aren't grasping. Apple has been developing AI in-house for the past decade. Rarely has Apple come out first on any technology unless it was completely developed in-house, and when it comes to industry wide products that change the course of the next decade they always hang back for 2-4 years while evolving their plans and cross-platform solutions.

When Apple comes out with their plans it will be more than what people are dismissing and in areas Nvidia has no foothold nor will they ever get a foothold in.

Nvidia had their Moonshot. That's Moon is starting to wane and they know it. Their valuation is not sustainable over the long haul and they know it.

They are nothing in the IoT world where hundreds of vendors are tuning their upcoming product lines with AI SoC across all network hops, across metropolitan to rural zones.

They don't know the consumer desktop, they don't have a shot to touch the iPhone/iPad/MacBook Air/Pro/Mini worlds, never mind the Mac Pro as they have no hardware in those tiers.

They are steadily falling behind in the Supercomputer space, they aren't a hub of Cloud Services like AWS, Azure, Oracle, IBM Watson, etc.

Apple will continue to diversify, extend and grow new vertical markets in the Smart Home, IoT markets, Streaming Media Services, etc. adding more to Apple One.

The Apple CarPlay research has barely scratched the surface.

CUDA is no longer the be all end all for Machine Learning, Diffusion, etc., and it's proprietary approach is why it will lose to AMD and others who are more and more investing in AMD gear each week.

Go try out TensorWave or LaminiAI.

Apple isn't betting their AI platform on the Client. Their heavy lifting will include the Cloud.

No offense but Apple also still can't make siri worth a damn and the things you're saying here people have been echoing for over a decade since the introduction of G-sync and DLSS. Everyone said G-sync was too expensive and wouldn't survive and DLSS is a pain but they're both superior products so they've lasted.

 

[caioferrari]

Maybe they should not release every new year a new soc and therefore cripple functionality. Looks like M3 got the old NPU from A15/M2 and not from A17 only to let the M4 look better.
Not much performance gain in CPU and GPU speed expected with the switch from N3B to N3E. So they needed the NPU thing.
Ah and I bet M4 pro gets the better Memory bandwidth again.

You didn't need to upgrade your M1 Mac when the M2 was released. The same goes for the M3. Remember that in this evolution, we had setbacks and then improvements and setbacks again with the NAND chip, we had memory bandwidth gains, support for ray tracing, and a video decoding core. You gained performance, but no processor was prevented from running any feature.

 

[FlyingTexan]

Yes, Siri seems to have been a dead end that turned out to be nearly impossible to modify and that Apple didn’t bother replacing for too long. It sounds like that is going to change.

Siri is not the only thing Apple has been doing with AI. They have been doing what they call machine learning for years, which is a functional AI focuses on specific features. Last year they did an experiment with the typing suggestions by using a transformer AI to learn your typing patterns. A transformer is what an LLM is but on a smaller scale.

I imagine an arrangement where Siri has a local LLM that listens to your requests (or via typing) and interprets the intent, even asking clarifying questions. Once it knows that you want, it would check its available functions in the OS or apps or web and bring those together to satisfy your requests. They could expand on the intents frameworks that expose app functionality to Shortcuts to provide a toolbox for the local AI. A lot of apps will have their own machine-learning-style AI functions, as well that could be called by a Siri. For some things, it will make more sense to pass the request on to a server-based AI like Gemini, OpenAI, or Baidu. This whole idea of the local AI that connects to other systems for different functions is similar to the Large Action Model envisioned by those crazy kids that made the Robot R1.

I’m really interested in where this could go.

Apple is so far behind they're in talks to use Google's system, which, honestly isn't very good.

 

[1984]

If it ain't broke, don't fix it. No need in change just for change sake.

They need to get rid of the notch. Somehow their competitors are able to put a camera in a thin bezel but Apple requires a notch and yet it doesn’t even have Face ID so it’s a design compromise for no reason.

 

[Realityck]

Last year, Apple introduced the M3, M3 Pro, and M3 Max chips all at once in October, so it's possible we could see the M4 lineup come during the same time frame. Gurman says that the entire Mac lineup is slated to get the M4 across late 2024 and early 2025.

Subject to where Gunmans moves his dartboard if wrong. Take the iPad Pro rumors, coming March, then late March to April, now delayed to May, soon it will be what’s likely to be announced at WWDC 2024. Apple these last 2 years has been difficult to approximate when updates actually happen against a lot of their product line outside iPhones.

 

[seek3r]

Maybe they should not release every new year a new soc and therefore cripple functionality. Looks like M3 got the old NPU from A15/M2 and not from A17 only to let the M4 look better.
Not much performance gain in CPU and GPU speed expected with the switch from N3B to N3E. So they needed the NPU thing.
Ah and I bet M4 pro gets the better Memory bandwidth again.

If the machine you bought does what you need it to how is it crippled?

I do my *job* using an M1P Macbook Pro, it works just fine, it didnt stop working because the M3 SOCs are faster.

Also, my dude, either you’re not old enough to remember or have bad enough memory to forget the massive year-on-year advances of computers in the late ‘90s and early aughts. The difference between M3 and the future M4 has nothing on that era.

You buy when you need to buy, there’s always something new around the corner.

 

[seek3r]

No offense but Apple also still can't make siri worth a damn and the things you're saying here people have been echoing for over a decade since the introduction of G-sync and DLSS. Everyone said G-sync was too expensive and wouldn't survive and DLSS is a pain but they're both superior products so they've lasted.

Tbh Siri does mostly what I need it to - set timers, control my house lights, set reminders, pause the appletv, find my phone, and do a bit of math or unit conversion every so often. Sometimes I use it to define or spell a word.

I dont really need it to be a genius in a box

 

[caioferrari]

Subject to where Gunmans moves his dartboard if wrong. Take the iPad Pro rumors, coming March, then late March to April, now delayed to May, soon it will be what’s likely to be announced at WWDC 2024. Apple these last 2 years has been difficult to approximate when updates actually happen against a lot of their product line outside iPhones.

The MacBook Air M3 was released one year after the predictions.

Now, TSMC has started announcing that it's ready to produce 2nm chips. In other words, for those who live on rumors, they shouldn't be swapping computers anytime soon.

 

[caioferrari]

If the machine you bought does what you need it to how is it crippled?

I do my *job* using an M1P Macbook Pro, it works just fine, it didnt stop working because the M3 SOCs are faster.

Also, my dude, either you’re not old enough to remember or have bad enough memory to forget the massive year-on-year advances of computers in the late ‘90s and early aughts. The difference between M3 and the future M4 has nothing on that era.

You buy when you need to buy, there’s always something new around the corner.

Sometimes I think rumor websites do a lot of harm to anxious people.

One thing is to be upset that the new computer was released the week after you bought a new one. Another thing is to be upset about the rumor that the company will release a new processor a year after your purchase.

 

[ThatGuyInLa]

Studio, Mini, and Pro skip M3?

 

[Biro]

Studio, Mini, and Pro skip M3?

IF Gurman is right (and that’s a sizable if) this scenario would be very likely. Remember, the iMac skipped the M2. So there is precedent.

 

[Chuckeee]

Do you guys think it would be a mistake to buy right now an M3 Pro Max with 15% discount or it would be smarter to wait for the fall and the release of this new M4?
Thanks.

There is always something newer and better around the corner.

The traditional adage (which is still true) is to wait until you actually need it, don’t upgrade just because you want it.

 

[Torty]

If the machine you bought does what you need it to how is it crippled?

I do my *job* using an M1P Macbook Pro, it works just fine, it didnt stop working because the M3 SOCs are faster.

Also, my dude, either you’re not old enough to remember or have bad enough memory to forget the massive year-on-year advances of computers in the late ‘90s and early aughts. The difference between M3 and the future M4 has nothing on that era.

You buy when you need to buy, there’s always something new around the corner.

I remember the 90’s but I didn’t had a Mac. So yes there were new processors all the time but I was always able to run all the apps. For sure not as fast as the newest CPU but I could run it.
I never experienced an app which I could not run cause I didn’t had the latest CPU.

But no so with apple. E.g. I could not run the latest games (resident evil) on my 14 PM one year later. Those were apple promoted games so it’s their strategy.
Also there are iOS features which don’t run on the last years flagship device.

That’s the reason why it’s not good if the cpu cycle is too fast or maybe it’s only me who dislikes this strategy and I should check for alternatives.

 

[transpo1]

When the M3 was in development, there was widespread anticipation for the new 3nm chips, with expectations of significant performance enhancements. However, upon its release, the improvements were only marginal.

Considering the AI capabilities of the latest processors, it's worth noting that the M series has had neural cores since the M1. Surprisingly, there hasn't been much discussion about performance boosts in NPU cores until now. The M3 showed a 15% increase in NPU speed compared to the M2, indicating progress, but still falling short.

The iPhone 15 Pro boasts an NPU nearly twice as fast as the M3, highlighting the potential for further advancement.

I believe that AI will become the primary selling point in the future, with companies focusing on NPU performance over CPU or GPU. Benchmarks will likely emphasize the speed at which devices can handle AI tasks. The M4 could potentially see significant enhancements in NPU performance, possibly through additional cores or improved speeds.

Is this a game-changing feature for you? If so, it might be worth waiting for the M4. It's important to remember that the AI competition is just beginning, and we can expect numerous improvements in the coming years. However, it's also worth considering that waiting indefinitely may become a perpetual strategy.

I anticipate some AI features will work with the M3 series and older chips, but they will of course run better and faster on the M4s with improved Neural Engines. I don’t think Apple will leave their older processors completely without AI capability.

 

[Realityck]

I anticipate some AI features will work with the M3 series and older chips, but they will of course run better and faster on the M4s with improved Neural Engines. I don’t think Apple will leave their older processors completely without AI capability.

I think what missing in this discussion is examples of all this speculative AI improvements can be applied to something consumer can use?

So what if you have faster neural processing in ARM SoC's, with only some limited examples of AI usage in your current apps and device usage without the more important software/OS abilities its just kicking the can some more distance, not creating a must have product. Apple has currently several product line situations that that thought applies.

 

[caioferrari]

I anticipate some AI features will work with the M3 series and older chips, but they will of course run better and faster on the M4s with improved Neural Engines. I don’t think Apple will leave their older processors completely without AI capability.

"I would be worried if the M1 - M3 didn't have an NPU and the M4 did. Now, worrying that Apple will focus on NPU in the new chip is as crazy as worrying that next year's computer will have a faster CPU."

 

[seek3r]

I remember the 90’s but I didn’t had a Mac. So yes there were new processors all the time but I was always able to run all the apps. For sure not as fast as the newest CPU but I could run it.
I never experienced an app which I could not run cause I didn’t had the latest CPU.

But no so with apple. E.g. I could not run the latest games (resident evil) on my 14 PM one year later. Those were apple promoted games so it’s their strategy.
Also there are iOS features which don’t run on the last years flagship device.

That’s the reason why it’s not good if the cpu cycle is too fast or maybe it’s only me who dislikes this strategy and I should check for alternatives.

I wasnt talking about Macs, in fact apple hit a bit of a wall on the power consumption of PPC chips for a bit towards the end there, which is why we never saw a G5 powerbook, just iterations of the G4.

As for modern obsolescence… does it look to you like Apple is rapidly deprecating AS machines?

 

[caioferrari]

I think what missing in this discussion is examples of all this speculative AI improvements can be applied to something consumer can use?

So what if you have faster neural processing in ARM SoC's, with only some limited examples of AI usage in your current apps and device usage without the more important software/OS abilities its just kicking the can some more distance, not creating a must have product. Apple has currently several product line situations that that thought applies.

Yes. What's being said is: Apple will bring NPU improvements in the new processors. It's likely that the new systems will start to feature AI capabilities, making it a benchmark for future devices. It's known that the M3 has 60% more NPU performance than the M1 and 15% more than the M2. Have you seen anyone caring about this? I haven't. It's likely that from now on, this concern will exist. It's only natural that the company invests in increasing/improving the NPU cores just as it has done with GPU and CPU cores in recent years.

 

[transpo1]

I think what missing in this discussion is examples of all this speculative AI improvements can be applied to something consumer can use?

So what if you have faster neural processing in ARM SoC's, with only some limited examples of AI usage in your current apps and device usage without the more important software/OS abilities its just kicking the can some more distance, not creating a must have product. Apple has currently several product line situations that that thought applies.

Apple is smart enough not to launch improved AI capabilities without capabilities that consumers will use. There have been recent reports of them working on AI photo generating capabilities, etc. I predict there will be some pretty useful and impressive capabilities built right into the stock Apple apps in iOS and macOS, especially photo/video/Siri.

 

[Jumpthesnark]

Guess mini skips M3.

That was my thought too. I'm kinda bummed, I had fingers crossed for an M3 Pro mini in a couple months.

 

[Ad47uk]

Makes me glad I got my Mac mini m2 pro last year, it means I don't have to buy another machine with AI in.
this AI stuff is getting too much, fine to have it in software, it can be removed, but not in hardware
I can't see myself getting another computer for a long time and at the age of 59, could be never.

 

[richinaus]

Apple is smart enough not to launch improved AI capabilities without capabilities that consumers will use. There have been recent reports of them working on AI photo generating capabilities, etc. I predict there will be some pretty useful and impressive capabilities built right into the stock Apple apps in iOS and macOS, especially photo/video/Siri.

I agree and imagine Apple Ai just to be implemented within apps and be seamless for users. It won't be an obvious thing.

 

[smulji]

Apple isn't betting their AI platform on the Client. Their heavy lifting will include the Cloud.

If recent reports / rumors are true, it looks like Google will provide the heavy lifting on the Cloud side of things.

 

[dgdosen]

What does AI-optimized mean ? Faster memory access ? more processing cores ? new intercores communication channels ? different cores ? less floating point, more fixed point processing power ?

I'm assuming it's with regard to inference (which I didn't see mentioned on this thread yet) on an existing model (vs training a model)

Apple missed the start of the AI race, and now they're scrambling... Sure, they can and should make a bigger "Neural Engine" in their chips, but I wonder which iteration of M hardware will be able to get that focus - because whatever penny pincher forced the 8GB as a standard was probably against increasing the budget for more neural cores when they were designing the M4 chip.

If Apple wants to be the 'inference' champ they could have hinted that - but they haven't. WRT memory - it's a shame they've been so stingy on the 8GB issue because AI calculations love to eat memory. It will be truly ironic if they introduce some offering at WWDC and it literally chokes on any base model laptop with 8GB. You know there will be active comparisons between a Mac running SiriV2 vs a Windows laptop running Copilot.

Maybe they'll expose/use more API surface on their GPU/NPU and tell us more about it at WWDC. I can see the idea of using your iCloud content to personally fine-tune some other model - but that's beyond my depth - especially from a security point of view.

We'll see more at WWDC!