Probably need to wait for M4, states my internal rumors
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.
Macs to Get AI-Focused M4 Chips Starting in Late 2024
- Thread starter MacRumors
- Start date
- Sort by reaction score
I’m guessing the iMac and the Mac mini will alternate to skip 1 generation.
If you were a true Apple fan, you know they gonna use it for smoother Memojis.
Same here, I use a dark background by default which makes the menu bar black and the notch therefore blends in.
From my understanding Face ID requires a prism for the dot projector which is too thick to fit in the lid of a laptop; the dot projector assembly is almost as thick as an iPhone or iPad. Optics are one thing that is rather difficult to miniaturize. The notch doesn't help with the fact that more depth is needed.
Don't forget the scratched up screens during production of all Pros.
I was just reading the other day about a new approach for this functionality that doesn’t require a VCSel (probably have the acronym wrong) laser or a prism to accomplish the task. I expect the Android and Windows/MacOS worlds will finally being seeing FaceID equivalents in the next year or two.
That "Unpatched Security Flaw" does not affect M3 users. [Confirmed by security team]. Only M1-M2 users.
GoFetch exploit can't be disabled on Apple's M1 and M2 chips
For now, cryptographic work should be run on slower Icestorm cores
am i the only person who sees AI plastered all over a product and wants to buy it less?
I'm all for new chips, but what the Mac's need more than anything is FaceID, Cellular and OLED/MINI-LED across the lineup where appropriate.
So there are 2 phases for modern style AI (AI isn't a single thing but a collection of techniques, most people mean deep learning or machine learning nowadays). Anyway the 2 primary phases are training a model, which is insanely energy and compute intensive and whatever machine you have isn't big enough (because if you build it bigger, well then you just add more parameters to your model! When I do some model training I use all 24 cores on my Intel box along with the +11,000 CUDA cores on my NVIDIA GPU since ML training is a massive matrix operation and GPUs are massive matrix crunching cards), but once a model is trained it takes, way, way less horsepower to use the model to classify inputs; think of your brain, learning to do a complex task (say surgery) is a very long intensive process, but once you've learned it, performing surgery as a trained surgeon is much much easier [as a doctor I speak from experience]. Unlike say our brains, most AI models are fixed once trained, so they are not continuously learning from their ongoing use, some do, but most don't which greatly reduces the compute load on the end user's computer. Another load folks often ignore (well not the people doing it) during the training phase (other than the brutal matrix operations) are the feature selection, which is where the algorithms pick out the important distinguishing features of the training data. For instance think of teaching your kids how to tell the difference between a cat and a dog, I mean in many features they're pretty similar (quadrupeds, furry, sharp canine teeth, good hearing, good sense of smell, claws) but clearly they are different species, and you could select features such as diet (cats are obligate carnivores, while dogs are omnivores, but more carnivorous), dogs cool via panting while cats cool by stealing souls via cold stares, etc, etc. These tasks may take more or less computation to pull out, and so sometimes the hardware acceleration is on the feature detector.
So the LLM models we all see in every marketing hype post took an arctic ice cap melting(tm) amount of compute to create, but you can use many of the generative AI systems on a hefty desktop to create content on your machine. As you can imagine with many algorithms, you can find steps where you can design specific hardware to do the steps faster than a general purpose computers (such as using GPUs in the training part which is kind of a "misuse" of their intended purpose), so Apple is jamming a bunch of hardware functions that their AI libraries need to do the classification step. A lot of AI code assumes a lot of parallel processing capability (what you need to do big matrix operations). there are of course other steps which are possible to accelerate in the pipelines (on both the training and classification side)
If you want the big forehead back then just use this and stop whinging.
And menu items gets hidden under notch. It so stupid.
There simply should be no notch since there is no Face ID.
It will get no Dynamic Island because Mac’s have no Face ID.
It is an incredible waste of screen space.
It's only a name though - you only think it's "almost too late" because you're thinking about the number in the name.
In terms of performance it's by far the lead processor until the next Ultra comes out...which will be 18 months later, the name is irrelevant.
Well. The Macbook Air can be configured to exactly 24GB of RAM.
Jony Ive gave us the butterfly keyboard and 1 USB-C port. No thanks.
wait, my m1 supposedly came with some 16-core neuter engine.. what happened to that??
Dynamic Island is so bad it's preventing me from ever buying another iPhone until it's gone. Notch and Space Black are preventing me from buying a MacBook Pro. If they punch a slot into a laptop screen just to accommodate someone's bright idea then they are officially dead to me forever. Your move Apple.
Thought I was going to upgrade from my M1 Ultra to an M3 Ultra this year, but I probably won't be able to afford it.
This makes me hopeful about eventually saving up and moving to an M4 Ultra (hoping it'll be released in spring 2025)
This makes me hopeful about eventually saving up and moving to an M4 Ultra (hoping it'll be released in spring 2025)
More like how dare Mark Gurman blab the M4 is coming soon, and new M3 Macs are not even out.
