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.
M4+ Chip Generation - Speculation Megathread [MERGED]
- Thread starter GoetzPhil
- Start date
- Sort by reaction score
Tim Millet said “we have one more chip in the M4 family to introduce“ which is a decent indicator. I think the Air will get M4 unless they got N3B costs down, and it will give them more margins since it’s such a mature design already being produced.
I’d say odds are 40/60 that Mac Studio and Mac Pro do something interesting with their next CPU. The Mni going up to 64GB suggests to me the Studio isn’t coming for a while, but who knows, they did release M2 Ultra 3-4 months before M3 but that processor was also never going to scale up.
Looking forward to seeing what the interconnects are once we get die shots of the Max.
If the Mac Pro is getting a chip with a different architecture from the M4 - would it make any sense to call it an M4 Extreme?
Apple could hardly call the new architecture M5 because it would never find its way into a power (wattage) conscious mobile device.
Calling the new architecture an M4 would also ‘date’ the architecture and imply that it is not as advanced as an M5 when it comes out a few months later.
My guess is that Apple will come up with a new letter at the front of the Mac Pro ‘extreme’ chip - like the D1 (Data Centre) or P1 (Professional).
If Apple considers the users of Mac Studio to also be professional Users for which the Mac Mini was not sufficient - it could lead to a marketing dilemma if Apple don‘t offer a version of this in a smaller, ‘snappier form factor like the Mac Studio.
I speculate that the Mac Studio will not get the M4 Max in mid 2025 … but will have entry level P1 chip that will outperform the M5 Max when it comes out in 2025 Q4.
Apple could hardly call the new architecture M5 because it would never find its way into a power (wattage) conscious mobile device.
Calling the new architecture an M4 would also ‘date’ the architecture and imply that it is not as advanced as an M5 when it comes out a few months later.
My guess is that Apple will come up with a new letter at the front of the Mac Pro ‘extreme’ chip - like the D1 (Data Centre) or P1 (Professional).
If Apple considers the users of Mac Studio to also be professional Users for which the Mac Mini was not sufficient - it could lead to a marketing dilemma if Apple don‘t offer a version of this in a smaller, ‘snappier form factor like the Mac Studio.
I speculate that the Mac Studio will not get the M4 Max in mid 2025 … but will have entry level P1 chip that will outperform the M5 Max when it comes out in 2025 Q4.
No, it wouldn't. We discussed this at some length earlier in the year. I was inclined to your point of view, but I was wrong. The Max is around 550mm^2, the reticle size is ~862mm^2. It is easy to imagine them being able to fit the whole 2xMax in, if you allow for removing the UltraFusion and some other redundant parts that aren't useful in an Ultra.
But that would be a truly giant chip. I think they'd rather build a maybe 1.7-1.8x Max monolithic chip. That should generally perform roughly on par with the Ultra, given the significant improvement of not having to route data over UltraFusion when the chip is busy (which is when it matters). (Perhaps it would do a bit less well compared to the Ultra in truly embarrassingly parallel code, but no matter.)
Then double that up with UltraFusion v2, getting you an Extreme worthy of the name.
You're almost right. They won't dump the E cores because they're useful and very small. A 4E cluster would have a negligible inpact on die area.
Also, I wouldn't be surprised if they soon start segregating a small cluster of E cores (or, radical thought, something like 1xP 2xE, though that is probably too hard/large) for the OS. That would remove a large class of vulnerabilities entirely, and make for more predictable performance.
As others have already explained, this is his Hidra rumor. If we see no UltraFusion on the M4Max die shot, that will make it likely that he's right. If we do see it, he's definitely wrong.
RAM size is at Apple's discretion. They could easily offer 256GB on the Max right now.
You're missing the Rev B Blue and White G3 that came out in June 1999, a whole two months before the PowerMac G4.
Thanks, that screws up the whole data set 😂
Many of us Rev A buyers felt pretty screwed at the time.
Last edited:
OK, so we're dredging up data from 1999 to argue about update rates in 2024?
I'm out. This is not a serious discussion.
I think we are still in the early days of all of this, but I can see the Mac Pro and Studio being put on a biannual update schedule. I expect Apple to spend a couple more generations of Apple Silicon tweaking and altering silicon layouts before we can reliably predict core counts and other components.
I do hope that the Ultra is a monolithic design with the ultra fusion interconnect so that we finally can have the Extreme SOC. That could make the wait worth it for some of those who didn’t get the M2 version.
The crazy thing is that the update cadence for the Pro did indeed drop off considerably after Steve Jobs passed away. If you look at the thirteen years prior to his passing there are seventeen updates compared with four that happened afterwards.
I do hope that the Ultra is a monolithic design with the ultra fusion interconnect so that we finally can have the Extreme SOC. That could make the wait worth it for some of those who didn’t get the M2 version.
The crazy thing is that the update cadence for the Pro did indeed drop off considerably after Steve Jobs passed away. If you look at the thirteen years prior to his passing there are seventeen updates compared with four that happened afterwards.
Last edited:
That's interesting because I don't think I've either not come across that part or paid close enough attention to what he meant with that rumor. But it sounds like that's on me.
In any case, the other concern I had that cast doubt on them being able to make a monolithic Ultra was the feasibility of it on future process nodes, where apparently reticle size limits had to be cut in half to make further progress on density with EUV double-patterning. But now it sounds to me like they want to rely on other optimizations instead to drive forward the density for a long as possible after hearing this from an SVP of TSMC:
“I hope we don’t have to reduce the reticle size because what we see is people want to integrate all functions closely together.”
Agreed, I think we’re currently still on a transitionary period where they’re laying the groundwork for being able to scale up higher and sync the generations of hardware every device uses faster (and what I mean by that is I expect to see fewer differences between things like the WiFi and Bluetooth generations that different devices use when they start putting their own modems inside or a few years into that transition (although things could go even further the opposite way if they fail massively on the project).
And if they want to implement newer ideas again in the future, they'll go into a couple years of a transition period again and then rinse and repeat the same process.
The M1 and M2 generations both debuted on the MacBook Air.
When Apple first launched Apple silicon with m1. I was so confident that M would be Mobile, rather than apparently Mac. I was also sure they would use D (Desktop in my thinking) or (P power/professional) etc.
It just makes more sense to have a chip that's not so focused on power efficiencies and e cores for these desktop machines. But i think the fact that the market is now so heavily portable focused, really makes the investment into any such big departure questionable. Perhaps less so if they need it for their own cloud compute infrastructure.
I feel like this is a bit of a cop out, they showed data from 1999 to present. As they say only really Tim era has it had slower cadence.
However, in that time we have seen a significant shift in the market to portable machines as they have become more capable. I also question how significant some of the updates in their list is.
The glory years of frequent Mac Pro updates were when Apple got to ride Intel's Xeon launch cadence, which was subsidized by Intel's much higher Xeon sales volume in the PC workstation and server markets.
Then, Apple made some big product design mistakes with the 2013 Mac Pro, and seemingly couldn't admit it for several years, resulting in essentially no meaningful updates. As they ended up admitting, they designed themselves into a thermal corner, one not well-served by the workstation CPUs and GPUs Intel and AMD wanted to make. If the 2019 MP hadn't launched just one year before Apple Silicon, I bet they would've returned to usually keeping it refreshed with the latest Xeon workstation chips.
Now Apple's entirely on its own, and while it's been a huge success overall, for high end desktops specifically they're in a tough spot. The 2013 MP malaise era cost them a lot of workstation Mac customers, and the Mac market is still quite small relative to the x86 PC market, so it may be difficult to justify taping out chips specifically for high end desktop Macs. It looked like Ultra Fusion was the answer in M1 and M2, because with that scheme they got to share tapeouts with the 'Max' chips, but UF went away in M3 Max and we're not yet sure about M4 Max. We should learn more next spring, and perhaps that will come with a more permanent answer for how they plan to continue designing silicon for high end desktop Macs.
Single Core:
M1 Max 2374
M2 Max 2748 (+%15)
M3 Max 3107 (+%13)
M4 Max 4060 (+%30)
Multi Core:
M1 Max 12229
M2 Max 14676 (+%20)
M3 Max 20927 (+%42)
M4 Max 26675 (+%27)
M1 Max 2374
M2 Max 2748 (+%15)
M3 Max 3107 (+%13)
M4 Max 4060 (+%30)
Multi Core:
M1 Max 12229
M2 Max 14676 (+%20)
M3 Max 20927 (+%42)
M4 Max 26675 (+%27)
One thing perhaps missing from that timeline is the Retina 5K iMac and iMac Pro — I mean, I spent about $4000 for my second one (I actually had two, a 2015 and a 2017), which replaced a Mac Pro. So I don’t think the urgency to rectify the errors with the 2013 Mac Pro was there in 2015, and by 2017 there was the iMac Pro.
The other thing I think we tend to underestimate is the impact of Apple’s profound, orders-of-magnitude transformation in scale during this period, basically going from having millions of customers to having billions of customers. The business decisions involved are different, there’s something sort of impersonal about operating on their current scale that is difficult to gauge. That’s what’s so annoying about the “Tim Apple” comments, the failure to grasp the runaway-train aspect of Apple’s transformation after changing the world with the iPhone.
I fully understand I may be a minority at this point but I don't want to pay for a small screen, batteries, bad speakers, worse camera, less than full size keyboard with numpad, trackpad etc. Annoyingly it seems that the desktop machines get passed over for generations and they start with the same mobile chip as in laptops. Tim has deemed that the big desktop machines just aren't as sexy in a spreadsheet. At least Jobs understood the point of a halo product and that it wasn't going to be their cash cow.
Just like every PC gamer doesn't own the Geforce RTX 4090 (less than 1% according to Steam which honestly is still surprisingly high).
If Apple updated the Mac Studio with the M4 Max I would definitely upgrade immediately seeing those performance increases (M2 Max to M4 Max single-core +49% and +81% multi-core).
Just like every PC gamer doesn't own the Geforce RTX 4090 (less than 1% according to Steam which honestly is still surprisingly high).
If Apple updated the Mac Studio with the M4 Max I would definitely upgrade immediately seeing those performance increases (M2 Max to M4 Max single-core +49% and +81% multi-core).
There is still some forlorn hope that M4 will introduce a dedicated foundation for the Ultra. Most likely not, and certainly there was nothing in the M4 Max presentation to indicate it, but Gurman’s rumor about a “variation” in the M4 family is still out there.
But yes, if it turns out that the entry-level Mac Studio is still the M4 Max, then I think it might require executive action to counteract whatever inertia is causing the real-world frustration @Pressure is articulating.
Perhaps Apple have made a mis-step on their desktop strategy?
M4 Max : Why force desktop users to wait until (rumoured) mid-to-end 2025 to get the M4 Max offered in their end-of-2024 140W laptop? Why was it not also offered in the Mac Mini - which is rated for 155W with no screen to drive?
How many sales is Apple loosing from Desktop 'M4 Max' holdouts between now when the Mac Studio refresh comes out? Will those users wanting a M4 Max then wait to see how the M5 Pro/Max laptops compare?
Perhaps what Apple should have done:
- Release the Mac Mini with the M4 Max now
- Announce that in 2025 H2 they will be moving their Studio and Pro to a new HPC that can scale to the most demanding profession needs.
- Set the base for the Studio to be (equivalent to) 1 x M4 Ultra - and the top spec 2 x M4 Ultras
It would keep the (M4 Max wanting) Desktop users happy in 2024 - and set the HPC professional minds at rest that in 2025 they can finally buy a machine that has sort of performance they've been dreaming about.
<edit : just seen tenthousandthings also responded>
Or large chips are simply harder to perfect (design, optimize, debug) than small chips? Xeons also tend to ship substantially later than their equivalent desktop/laptop chips.
It's "easy" to simply slap two chips together, it's a lot harder to get the result to run twice as fast, using only twice the power. There are endless issues, from how to distribute work to how to handle interrupts to power management to coherency.
The M1 Ultra clearly worked OK, and just as clearly did not work as well as one might have hoped. The M2 Ultra fixed the most immediate problems, but still left a number of issues unresolved. Meanwhile Apple probably wants to grow these things even bigger, at least to the next stage of four chiplets.
So what do you do? You can continue to ship adequate but disappointing products (like an M3 Ultra, say) or you can engage in some hard work to fix all the known problem areas, accepting that that will take as long as it takes.
I'm assuming the non-release of a Studio M4 Max implies that this "as long as it takes" actually means a release in the "near" future (maybe Q1 2025) rather than "we have no idea, maybe with the M5".
But I think this also means that Apple knows they don't get second chances. If they ship a disappointing Ultra or Extreme they will have to live with that for a decade. Better to delay a year of sales than take that hit.
Remember there are always other players in this game than the ones you know about. If Apple have the fastest single cores in the world, and can also pack those into a large enough throughput, that becomes interesting to clients you don't boast about but who can pay an awful lot - high frequency trading, various government agencies, ...
But you don't want to blow your once a decade chance with those customers by shipping an Ultra or Extreme that falls down (fails, or just slow) on some vitally important issue - remote atomics? cross-chiplet coherence? hypervisor? large pages (a new feature for Apple, but one that the highest end may well want)?
Is there any speculation on how they get 36 GB of RAM on M4 Max? With M3 it was 3 x 12 GB, but now it should be something like 4 x 9 GB, is it even possible?
It looks to me that 14 inch MacBook Pro motherboard only have group of 3 memory modules, while 16 inch MacBook Pro have group of 4 memory modules.
I think you mean the difference between the 410GB/s M4 Max and the 546GB/s M4 Max. There is no difference between the 14" and the 16" that limits memory bandwidth and capacity.
But yes, I’m sure that must be right. 36 GB memory is only available in the 410GB/s M4 Max (3x12), while the 546GB/s M4 Max starts at 48 GB (4x12)…
Last edited: