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

[ThatGuyInLa]

Ai performs "approximate" calculations for you quicker than ever before! Not perfect, but for most things it's great!
You've been living this way with Siri for the last 14 years! BUY NOW!

 

[novagamer]

That “unpatchable security flaw” is just clickbait. It requires the exact same chunk of data to be repeatedly encrypted and decrypted while the SoC is being physically scanned the entire time, which would never happen in real life. Especially because the decryption step can only happen if the password has already been entered!

This isn't true, no physical access to the hardware is required. If you are only familiar with the older attack(s) from last year I suggest you read about the newer methods, the research work has also been updated to specify which versions of MacOS were used and not Linux as many people were speculating in previous threads.

Apple's mitigations do not fully solve this for M1/M2 and for M3 they don't solve it unless you use their own libraries and / or structure your code in a specific way, check their updated ABI documentation.

There is a bit that can be flipped in EL1 for M1 & M2 but Apple has not acknowledged it publicly yet. It could provide an M3-like mitigation if they can trigger it during specific operations.

Please provide a source if you have updated information available, either pending review or peer-reviewed, I'd be very interested to read it.

 

[Darren.h]

DARN. M4 will still be on 3nm?? Looks like it is the M5 2nm I want. Just like the BMW M5.

 

[Yoda Mann]

Still gonna be a problem?

A recently discovered flaw in M-series Apple silicon chips could allow attackers to break Macs' security, according to researchers. “GoFetch” is an attack method targeting data memory-depending prefetchers (DMP). Apple's M1, M2, and M3 series chips all include this feature—and there's no way to disable it for M1 or M2.

 

[hans1972]

but the iMac and Mac mini will still start with 8GB memory

It's called scalability. Starting with 16Gb of RAM would reduce scalability.

 

[picpicmac]

Well then, should I put off buying a new Mac in May or June as I had planned?

Guess I have to wait until November.

Who wants yesterday's model?

 

[Jdhommert]

AI stuff from Apple probably gonna be like

“Introducing…..SirAI!”

Now she’ll be wrong and **** **** up before you even tell her to do it”

And we think you’re going to love it!

 

[NT1440]

A.I. time, huh? So all those un-upgradable 8GB models won't be joining the party in any way shape or form? 😅 What a shocker 😆

Now picture this, Apple has thrown out “common knowledge” in terms of how “AI” is done on windows, and instead has been including the hardware required for their approach since day 1 on the M1 chip. Then they had the gall to show how they could do it on iOS with less available RAM and even published a white paper on it that MR reported on. Wouldn’t that be something?

 

[pcunix]

Apple is not behind AI at all, wtf?

Apple dominates Machine Learning which is the foundation of the "AI".

You most likely mean that they may lack behind in the GPT or LLM

Thanks. I was about to say something similar.

Too many people think AI is a Chatbot. ML (machine learning) is the AI Apple (and not just Apple) has been doing for years now. They most definitely are not behind.

However, people also see Siri and wonder why it cannot understand commands and questions as well as a Chatbot can, so they say “Apple is behind”. It would probably be more fair to say “Apple hasn’t put enough AI into Siri.”

Give the research papers they’ve published about stuffing an LLM into flash, recognizing objects in images, and this month‘s paper on recognizing controls and app capabilities by reading the screen, I’m not sure they are behind in LLM. They just haven’t shown us much of it—yet.

I think even what is already under the hood would be much more impressive if Siri could understand us better and act on it. Hopefully that’s what we’ll see in June.

 

[Eric_WVGG]

geeze two months and now I gotta drop kick this M3 Max into the river

fun while it lasted!

 

[NT1440]

I have a funny feeling the A18 Pro will be M4 like. I’m still confused/intrigued regarding the 35 TOPs on the A17 Pro and what effect it truly will have.

iPhone/iPad users may see more benefits of AI

You’re already seeing it today in computational photography, automatic image occlusion, advanced skeletal tracking, on-device language translation, etc.

Apple has been doing “AI” for quite a while, they just haven’t jumped on the rebranding train until now as they were calling it (much more appropriately) Machine Learning (ML) even up until WWDC of last year.

This is why I laugh at the notion that Apple is “behind” in “AI” is brought up by people who clearly just get their technical knowledge from tech illiterate places like CNBC or Bloomberg. They’re behind..because most people have no use for a chatbot? They’ve been doing groundbreaking ML on their mobile chips for years now, on device, with features that competing platforms simply can’t do without doing it off device on a server. 🤷‍♂️

 

[JordanCautious]

WTF 512GB of Unified memory?!?! I almost fell to my knees in a Wendy's. Imagine the cost lmao. Impressive though

 

[fuchsdh]

Well, AI is definitely the buzzword du jour, but unfortunately it needs AI software to go along with it.

We’ll see how that goes. A revamped Siri running locally definitely has the potential to be an intriguing proposition. However I’m starting to think the hardware will be ahead of the software for a while yet - and that Apple will be relying on 3rd parties for their AI implementations almost exclusively.

I just haven’t seen a cohesive presentation of how all of this fits together and will benefit your average user.

It seems like this is just Apple rebranding what it's already doing with the Neural engine, which has been a thing and will continue to be. But machine learning and large language models and "AI" are the hotness, so they're just going to tout it and say it's "2x faster" when they stick more cores in it and call it a day.

You're definitely right that there's a lot of hype that isn't justified, but plenty of software that exists now does use this stuff for real work—on-device transcription, for example, has shifted even in the last five years where I used to have to send out for transcripts from Rev.com and now I can get 95% of what I need in a tenth of the time just on my local machine.

 

[andrewxgx]

Will the next MacBook Pro get a redesign it’s had 3 generations with the same look

and whats wrong with it?

next redesign is probably when they get OLED ready

 

[NT1440]

Still gonna be a problem?

A recently discovered flaw in M-series Apple silicon chips could allow attackers to break Macs' security, according to researchers. “GoFetch” is an attack method targeting data memory-depending prefetchers (DMP). Apple's M1, M2, and M3 series chips all include this feature—and there's no way to disable it for M1 or M2.

It’s mitigated on the M3 hardware.

 

[teknikal90]

Of course I’m buying a Mac Studio to work and make money…it serves no other purpowe these days…

what workload are you worried that the m2 ultra can't handle right now that you think the m3 ultra can?

 

[darngooddesign]

And menu items gets hidden under notch. It so stupid.

The menubar text items don't but the right side menubar icons can.

There simply should be no notch since there is no Face ID.

You can hide the notch by choosing a different screen resolution.

 

[NT1440]

DARN. M4 will still be on 3nm?? Looks like it is the M5 2nm I want. Just like the BMW M5.

2025 is the earliest anyone is going to be on 2nm, and that’s if scaling production stays on target without any complications.

 

[darngooddesign]

Still gonna be a problem?

A recently discovered flaw in M-series Apple silicon chips could allow attackers to break Macs' security, according to researchers. “GoFetch” is an attack method targeting data memory-depending prefetchers (DMP). Apple's M1, M2, and M3 series chips all include this feature—and there's no way to disable it for M1 or M2.

It's a problem if you let someone install an application and grant it permissions. Don't install untrusted software.

 

[Eric_WVGG]

Will the next MacBook Pro get a redesign it’s had 3 generations with the same look

Every Macbook going back to the aluminum Powerbook G4 has lasted four release cycles (roughly four years) between redesigns, so the 2025 M5 looks lot more likely.

(also that's when the rumors of OLED MBPs are pointing, and Apple tends to roll out new display techs with redesigns anyway.)

 

[ElectricPotato]

Now it's going to be AI everything because they don't know what else to market :/

An accordion... with only one button... wait for it... because it is AI powered. We live in a magical time.

 

[seek3r]

Because my M1 Max has the ugly notch which to this day I hate and the webcam isn’t even that good so they can change it

I've never understood people who complain about the notch on a mac, it sits in an area of the screen that is rarely used, and because of that it allowed for thinner bezels on the rest of the top of the laptop screen and the menubar to move up, which meant freeing up a pile of space on the screen below it. It's a good compromise, and you legit dont notice it when actually working or using the laptop

 

[smulji]

So the delayed iPADs can have M4 🤣

You laugh but I wouldn't be surprised at all if the iPad Pro did have M4.

 

[crsh1976]

I'm not digging this trend of constantly rushing new chips to the market like Intel and AMD do it in some very sketchy way to create/perpetuate this feeling of always being behind and buying new stuff (I'm looking at you Intel and your insane marketing noise pushing the latest gen of chips that is essentially the same as last year's but with a different naming scheme).

I get that Apple isn't done rolling out the M3 variants yet, but overlapping generations because they're so slow at updating all models in a short time span is probably what I would them to fix first rather than keep pushing newer chips right away.

 

[hans1972]

Too bad AI is still a meaningless buzzword. LLMs are just a very fast way of synthesizing information, but it isn't actually "thinking" and there's no way to verify that it hasn't been misled. GIGO still applies, there's no way going through more information faster does anything to figure out what's correct information and what's still garbage.

Maybe you should think about why it's called artificial intelligence.

There is no requirement for AI to be like a human or human intellectual processes. It's enough if it mimics the results of human intelligence.

And the field has existed for at least 60 years.