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 M5 Chip's Dual-Use Design Will Power Future Macs and AI Servers
- Thread starter MacRumors
- Start date
- Sort by reaction score
Yes, I'm not doubting that Apple would be first in line for N2. I'm just not sure if a) N2 will really be ready for mass production next year and b) if it is, if Apple will use it for the M chips or only for the iPhone (with M chips following 2026).
It's going to pain Apple to spend the extra ~$20 required per Mac to double the base RAM and storage on every device... the fact they will no longer be able to get away with pathetic amounts of both is perhaps the greatest thing about AI on Macs 😅
Likely a good challenge in TOPS per Watt. And the TOPS Per Watt numbers are very important when the largest data centers are constrained by power and cooling, not by chip count. And the ML TOPS per dollar might also beat Blackwells for Apple, as Apple doesn't have to pay any Nvidia markup when using wafers of their own design built on TSMC fab lines that they partially financed.
Has anybody confirmed whether these iPads actually show 12GB, or if they've been knobbled to 8GB?
Apple can probably get away with that on the iPad - where the RAM size isn't advertised prominently and they don't offer BTO RAM upgrades separate from storage bumps. I think it would go down like a lead balloon if they tried it on a Mac and then offered a $200-for-8GB upgrade.
If I recall correctly, the 1TB+ iPads offer 16GB of RAM, not 12GB. If they're planning to drop the 8GB Mac RAM. If they're going to offer 12GB and 24GB Macs I don't really see a niche for a 16GB SoC. So the "temporary freebie thanks to price fluctuations" sounds more plausible.
Still, it's high time that the Macs went to 16GB minimum (except maybe for the cheapest MBA, which - when M4 comes out - will probably be the current 8GB M3 model anyway) especially since the new Copilot+ PCs (with ARM-based SoC processors) start at 16GB.
Yeah, I think even the cheapest of the cheapest machines should be shipping with 16GB minimum today. The Pro's should start at either the odd 24GB or 32GB preferably. I think even 12GB on the base model stuff is just OK, but not really enough of an increase, even if the AI stuff eats 4GB, then we are still effectively left with 8GB for all other things which is just where we are now that isn't enough. I've been on the 8GB starting is dumb train for the last 3 years or so. I didn't even like it on the M1 for longevity sake, but now it's just absurd.
M5 or M6 for me… I’m on my way to Birmingham! Ahahahahahha.
It appears to me that AAPL's strategy is to avoid owning capital assets and hiring people in its supply chain operations where it can. This gives it flexibility and mitigates risk by keeping hundreds of thousands of employees in jurisdictions across the globe and hundreds of billions of dollars of assets that can become obsolete, off the books.
AAPL is able to do this at a cost and quality level most other manufactures cannot achieve with their own facilities because of the scale at which they operate and the level of effort they invest in managing their partners and the processes. This "operational art" is the value that Cook brings and a significant reason why AAPL has a $3 trillion market cap.
So, because the chips are stacked on top of each other rather than on a long roadway to the edge of the chip die, they can communicate quicker? I know when I was working on the drum machine there was a clock signal line (wire) that went all the way around the edge, a huge, long distance, problem, and it actually did cause a problem! The board’s performance was flaky. And it seems silly to even think that, because electrical signals travel at the speed of light, right? But… Distance is distance. Trust me on this one.🍸😹🙀😹
Not really. My M2-Pro with 16GB RAM is running a local copy of Meta's Open Source Llama3 8b at a very decent speed of 28 tokens per second and that is while also running the usual Mac desktop, Chrome browser, and some other stuff.
I'd say my Mac is keeping up with an Nvidia 2070 GPU, The 2070 might be a little faster for running AI models but it is close enough. The better spec'd Macs can be up to 4X faster than the 2070, for AI. (I have no interest in Video and games, I'm only comparing AI tasks.)
So I have to disagree and say Apple did quite well with cost vs performance on AI tasks. The M4 is likely much better than my M2 for this kind of work. I'd upgrade except that it is my limited brainpower holding back the development of a robotics project, it is not yet the M2 hardware that is the bottleneck.
I might be at the point where the M2 is "slow" about the time the M5 is released.
One More Thing: Read the comments and doc that were released my Apple with thier open source OpenELM. IT says they are using Linux to train the model (We don't know if it Linux on Intel or Linux on Apple Silicon) because they say "Slurm is not yet running on MacOS." So apparently as of a few months ago, Apple had not yet built out a MacOS-based AI training farm. I'd guess it was Linux on Intel with Nvidia GPUs. But Apple's OpenELM does run on the M2 and even on an iPhone. This is Apple's "edge AI" that runs locally on customer devices. It has been out for a while now. Google "OpenELM" to find it. You will see that it runs well, Apple is not so "behind the curve".
If they use Gen AI the way they're doing so far I think it's fine - it's mainly for fun things. Creating Emoji I thought was a genuis idea to be fair.
I'm not sure they'd ever have the very best image generating AI models either.
Light moves about one foot per nanosecond, give or take a little. Electronic pulses in wire move a bit slower than light in a vacuum. The problem is not the delay but gettinght same pluse to every place it is needed at the same time, so the parts stay in sync. It is this "clock skew" that they are trying to fight
I could be wrong but I don’t think we’ll see Apple release any kind of AI that would steal work from legitimate creatives.
Everything they’ve announced so far are either for editing your writing or photos, or generating purposefully gimmicky things like emoji and cartoonish/sketch-like pictures, nothing photorealistic.
It’s not writing an essay for you, it’s taking one you wrote & rewording it. The pictures it generates are not ones anyone would have paid to have made, and it has set styles so people will know it was AI generated. I think this is all very intentional
I like to see more articles discussing what Apple uses in their server farms. Something that isn't discussed enough.
Right now the AI is used for stuff that is mostly only entertaining. But the intented is to using it for all the normal stuff you do with a computer like "Siri, I want torepay is susan's email about the 34th street job. Tell her, "let's go with her option #2".
The AI then figures oput which email this is, reads it the find option #2 and then composes a reply and puts it on screen with both "OK" and "edit" buttons. Not entertaining, just basic business. This is what Apple is working on now.
If AI is done right, it will not seem like AI, work under the hood and do what you want. One of my goals, I work on would be a robot and you ask it, "I my UPS package here yet? Bring it inside when you can." The robot's AI wopuld lok for delivery emails or simply look at the outdoor security camera (ring?) and when the package is there, it opens the door and brings it inside and shuts the door. This not 100% briilent but still way more advanced then anything we have now. The abilty to be very simple-minded tasks like this would be revolutionary, folding laundry and placing dishes in the dishwaters, taking out the garbage, Taking grosheries out of the bag and putting them away, cleaning countertops and so on.. No great feat is intellectualism but still very useful. I figure people would pay the price of a car ($35K?) for this. IMO, the "AI" should disappear into the product.
But why would Apple want you to buy a Studio if they’re making so much profit off selling you memory and ssd? Are they making even more profit with Studios? I’d be interested in seeing what the numbers are.
Your M2 Pro is better equipped to run ai tasks than an m4, or even the m3 pro, because it has higher memory bandwidth. Remember when people were making up excuses for cutting back on specs that it doesn't matter in anything average users would do? Well, not true anymore. Reducing memory bandwiths - with 10 core m3 max too - was probably a deliberate step to make the cheaper chips worse at ai tasks. Even the M1 Pro beats the M3 Pro because of the bandwith, despite the M3 Pro having much better gpu cores. Don't upgrade.
In what benchmark or scenario does an M1 Pro beat an M3 Pro? And I'm on your side regarding my dislike of how they handicapped the M3 Pro in hopes of selling more Max chips!
Hi Cervisia - are there stats on this? I would like to understand the memory bandwidth issue better because I need to upgrade my MBP to Apple M-Class chips from my current Mac-Intel Chip. I don't want an AI chip. But I also don't want a chip that has been intentionally degraded in terms of memory bandwidth just so that Apple can reduce production costs. (I don't love Apple so much these days. Feels like they are cheap penny pinching Ebinezer Stooges. I miss the Apple that made it their business to constantly surprise and delight me.) I want to know what is that inflection point that I should look for to help me know when to upgrade. If M2 had the best memory bandwidth, what was that throughput? I may wait for M4 Pro / M4 Max / M5 / M5 Pro / MX to match or exceed that M2 Pro data point.
I agree. I mean original Siri is AI, but people don't think of it like that anymore. New Siri is basically what they made out original Siri would be - a way for you to say what you want in so many different ways and with so many different words you can just speak naturally and have it action your command.
Define "market"...
They will certainly make large machines for themselves.
Will they sell those machines to others? Maybe? But maybe via a special program where there's no real public sale, more like buying a nVidia DGX than like buying a Mac.
Or maybe they won't sell them, all you can do is rent small or large amounts of time, like AWS.