Even if it were ready, will there be enough production to satisfy the dozens of millions chips Apple would need?
M4 Rumors (requests for).
- Thread starter Jimmyjames
- Start date
Even if it were ready, will there be enough production to satisfy the dozens of millions chips Apple would need?
Oh yeah, the M1 (LPDDR4X) vs M1 Pro/Max (LPDDR5) split. I could see that. There are reports like this which show that even though LPDDR6 hasn't been certified there seems to be accelerated interest in it so maybe TigerRick is making some points
Given that I would say it is more likely that Apple will go with LPDDR5X/T/whatever but I could see Apple paying Samsung for early access to 6 before general availability for the M-series. That said, the A18s would almost certainly have to have LPDDR5X/T/whatever.
So Phoenix L/M are Nuvia cores ... apparently looking stuff up I'm not the only one confused about how they relate to Oryon. And I guess they will have efficiency cores? which weren't ready for the laptop chips just a few months earlier? Though if Gen4 is announced in October how long until it actually ships to in products to customers?
For the iPhones, very unlikely if certification isn't even until Q3 2024. For the Macs, maybe. Naturally I have no idea what Samsung's production availability is, but I wouldn't rule it out entirely like I'm basically doing for the A18.
Agreed if I were put in charge of designing a monolithic Ultra (and it's a good thing I am not), I would keep some. Assuming a 45+ TOP NPU but keeping core counts of the M3 Max the same for the M4 (which may not be the case naturally), here's how I would slim down the M4 Ultra from two M4 Maxes:
basically 75%, why? I dunno I just chose that.
CPU: Max: 12 P-cores / 4 E-cores -> Ultra: 18 P-cores / 4 E-cores -> Extreme: 36 P-cores / 8 E-cores
NPU: Max: 1 NPU -> Ultra: 1 NPU -> Extreme: 2 NPUs
GPU: Max: 40 cores -> Ultra 60 cores -> Extreme 120 cores
Everything else about 75% too. Since these are desktop-only chips, you can then boost clock speeds for some of these cores, especially the GPUs, to (partially? wholly?) make up for the difference in what would've been had the Ultra been 2x a Max (taking into consideration lost compute bandwidth due to the interconnect, so the GPU would only have to boost < 10%, maybe even 5%). That would be fun and make the high end desktops true monsters.
Edit: I forgot to add additional PCIe lanes for the Ultra for additional Mac Pro expansion.
Last edited:
casperes1996
Site Champ
- Joined
- Nov 29, 2021
- Posts
- 325
Aside from the large quantity of iPhones and thus higher production requirements, there's also just the fact that the phone SoC has smaller memory bandwidth requirements than if you need to feed a huge GPU like in an Mx Max
Though I don't know NPU requirements on that front
B01L
SlackMaster
- Joined
- Dec 1, 2021
- Posts
- 197
True, but I would think 16 E-cores might be too many...?
I have been saying Apple needs to move the high-end desktop Macs to LPDDR5X for awhile; this could allow up to 1TB of RAM in a Mn Extreme configuration, with 2.16TB/s UMA bandwidth...?
Seems like you are arbitrarily cutting core count as the chip size goes up, just because...? The compute cores (CPU/GPU/NPU) are not what need trimmed, it is the extra subsystems that would only be used in the most extreme (no pun intended) cases; I could see something like this:
- 64-core CPU (56P/8E)
- 192-core GPU (the GPU is the main area where NO cores need to be cut)
- 128-core Neural Engine (more NPU cores than less; Apple IS focusing on AI this go-around, after all)
- 1TB LPDDR5X RAM
- 2.16TB/s UMA bandwidth
- For the Mac Pro, keep the same PCIe slot arrangement, but with full bandwidth available to all slots at once (no apportioning)
The current rumor, which may be incorrect, is that the Mac Pro will tap out at 512GB.
My initial condition is that the monolithic Ultra is too big to be economically feasible to fab for a reasonable Studio price, combined with Apple’s preferred profit margins ofc, and that cutting subsystems alone makes too small a difference*. The math behind that comes from earlier in the thread but even just a priori there must be a reason why, financially, Apple went with ultra fusion rather than monolithic for the Ultra chip beyond the upfront cost of designing another die. The GPU and CPU are the two largest areas. Therefore I was theorizing about how I would dramatically reduce the die area by 25% or more while reducing the performance by as little as possible compared to the standard Ultra design with an interconnect. Now 25% is a little arbitrary but it does allow for some clean multiples of the current design which may not be that important since the M4 Max may have a different core count. 25% is also nice since I’m assuming, perhaps erroneously, that the M2 Ultra GPU’s remaining lack of perfect scaling, the M1 Ultra was worse, is due to the interconnect alone. Thus for the GPU clock speeds would not even have to increase substantially to make up the difference. The CPU I believe would but then this is a desktop part, Apple might not throw efficiency away completely and follow AMD and Intel off that particular cliff, but they can stand to increase clocks on a desktop only part. Thus they can mitigate lower core counts and die area cut to save money by increasing clocks on the P-cores and GPU cores.
Cutting the extra NPU may not be necessary depending on its die size and indeed on Apple’s design priorities.
This is all highly speculative built upon layers of assumptions, but given that set of conditions, that’s how I approached my design criteria.
N3X I don’t think will be necessary for Apple to increase clocks by the amounts I’m suggesting and shouldn’t be necessary for your design either. This should theoretically all be doable on N3E, which according to the rumors is what M4 will be based on.
Edit: *And remember we’re eventually gluing some fraction of two of these together to make an Extreme chip, that has to be economical too! That’s why I’m trying to reduce die size more radically.
Last edited:
Yoused
up
- Joined
- Aug 14, 2020
- Posts
- 6,840
- Solutions
- 1
Over on ars, they are relating that the Great Gurman says the Mac Mini will skip M3 and go straight to M4 this fall. I have not kept tabs on his recent record, but why the hell is this hot air getting headlines?
Also, in a bit of "sad" news, they report the death of the Z80, at least in its classical DIP stand-alone form.
Also, in a bit of "sad" news, they report the death of the Z80, at least in its classical DIP stand-alone form.
- Joined
- Sep 26, 2021
- Posts
- 6,318
- Main Camera
- Sony
I didn’t even realize they were still making the Z80’s. I assume they long ago switched these to CMOS on a somewhat reasonable process? Or were these things still nMOS?
First computer I ever touched was a TRS-80 model 1. Fun that derivatives of these things are still in use - I think my daughter’s calculator has one in there.
Yoused
up
- Joined
- Aug 14, 2020
- Posts
- 6,840
- Solutions
- 1
You realize that the Space Shuttle thingy was running 8086 processors right up to the end, some 13 years ago? They were finding it hard to get parts for. Seems kind of odd that NASA, home of Nerd Central, could not spend a million or two to upgrade the fight systems.
It looks like it’s $50 million for their next-gen system: https://www.theregister.com/AMP/2022/08/17/nasa_space_computer/
Jimmyjames
Elite Member
- Joined
- Jul 13, 2022
- Posts
- 1,094
John Gruber posted this article on the M4 and the expected cadence.
Making a Mountain Out of Molehill-Sized M4 News
“*The entire Mac product line is set for annual speed-bump Apple silicon updates*” is, as far I can tell, the actual story. Not “*Mac sales are in the tank and Apple is overhauling the whole product line to change its focus to AI.*”
daringfireball.net
I know Gurman has good sources but often makes mistakes. Gruber also has sources and often used to talk about the “little birdies” that told him things. His article mentions he expects the M3 Mini around May/June. He may just be guessing, but I wouldn’t be surprised if he’s correct and Gurman’s wrong.
B01L
SlackMaster
- Joined
- Dec 1, 2021
- Posts
- 197
John Gruber posted this article on the M4 and the expected cadence.
Making a Mountain Out of Molehill-Sized M4 News
“*The entire Mac product line is set for annual speed-bump Apple silicon updates*” is, as far I can tell, the actual story. Not “*Mac sales are in the tank and Apple is overhauling the whole product line to change its focus to AI.*”
RE: Gruber and this comment from the linked blog post...
Just kinda irks me how everyone goes on about the 2019 Intel Mac Pro and the 1.5TB maximum RAM it supported...
Said RAM was only available for the top-end CPUs, and was a by-product of the 6-channel memory bus...?
All Mac Pros until then maxed out at 128GB of RAM...
The point being, the 2019 Intel Mac Pro & 1.5TB of RAM was an anomaly, an outlier...
LPDDR5X RAM should be the next step for Apple, and a (theoretical/hypothetical) Mn Extreme "SoC", with a possible 16-channel memory bus, utilizing 64GB chips, should be capable of supporting 1TB of RAM with a 2.16TB/s UMA bandwidth...
Not 1.5TB of RAM, but 2/3 of the way there, and much better than 192GB...?
Just hoping WWDC 2024 brings us some real news/excitement regarding the high-end Mac desktops; these being the Mac Studio, the Mac Pro, and the possible Mac Pro Cube...! ;^p
EDITED - 2019 Mac Pro memory channel count & overall Mac Pro maximum RAM support...
Last edited:
Nycturne
Elite Member
- Joined
- Nov 12, 2021
- Posts
- 1,392
Yeah, the TI-8x line had them when I was still in college. Because we were already supposed to be learning assembler, I used mine in a couple CE projects with apps written in assembly. The eZ80 is still used in TI's current crop I believe.
- Joined
- Sep 26, 2021
- Posts
- 6,318
- Main Camera
- Sony
Neat. I was always an HP guy. RPN or die. But it’s fun to imagine the calculator with the same chip as all those early computers and game consoles.
I view it as all a game: it’s fun to talk about and fuel for speculation but no proclamations from anyone outside of Apple should be taken too seriously until made less than a few weeks before the claimed launch - basically anything said ~3 weeks from now for any WWDC launch (or lack thereof) in early June and ~early October for any possible M4 release. Of course WWDC launches (or no launch) and the iPhone launch will already give us some idea of what is coming (or not) by then for the M4.John Gruber posted this article on the M4 and the expected cadence.
Making a Mountain Out of Molehill-Sized M4 News
“*The entire Mac product line is set for annual speed-bump Apple silicon updates*” is, as far I can tell, the actual story. Not “*Mac sales are in the tank and Apple is overhauling the whole product line to change its focus to AI.*”
I know Gurman has good sources but often makes mistakes. Gruber also has sources and often used to talk about the “little birdies” that told him things. His article mentions he expects the M3 Mini around May/June. He may just be guessing, but I wouldn’t be surprised if he’s correct and Gurman’s wrong.
That’s a fair point though of course part of that was due to the infrequency of updates for the Mac Pro line (the previous update was 2013 I think? And that was the trash can Pro?). But yes they offered that config only once and was, if I remember right, incredibly pricey (even if you got the memory 3rd party).
My Dad is grimly hanging on to his last working RPN HP. Of course there are those HP emulators in the App Store …
The Space Shuttle didn't use the 8086. It couldn't, they began designing its avionics before single-chip microprocessors were even a thing.
It used IBM AP-101 computers, a miniaturized variant of IBM's 1960s S/360 mainframe architecture. According to this NASA history:
computer
www.nasa.gov
each AP-101 weighed 50 pounds and used 370W. It had five of these - four in a quadruple redundant setup as the primary flight control system (using voting to resolve differences in opinion), and a fifth running a different software stack (written by a different team, to somewhat different requirements, reusing zero lines of code). This was a kind of ultimate backup - in case some bug in the complex primary system with its redundancy made the flight computers unusable in a mission, they'd have an alternative to switch to.
And (as you probably know) even with the advent of single-chip microprocessors, NASA sticks to much older systems. AFAIK, NASA and others are still using the radiation-hardened RAD750 CPU for their most extreme applications (it was included in the Webb), which is made on either a 150 nm or 250 nm process—and costs ~$300k/unit!
They stick with it because they know it works, and its ~300 MIPS is sufficient for their rather modest local processing requirements. The large feature size is a benefit, since it reduces the ability of radiation to disrupt or damage the processor.
RAD750 - Wikipedia
Last edited:
True. But I thought those complaining the max RAM on the AS MP was too low weren't saying it was too low for a Mac (where, as you say, 1.5 TB RAM was an outlier), but rather that it was too low for a workstation (where 1 TB - 2 TB max RAM is typical), since the MP is supposed to be a workstation-class machine.
Last edited:
- Joined
- Sep 26, 2021
- Posts
- 6,318
- Main Camera
- Sony
And (as you probably know) even with the advent of single-chip microprocessors, NASA sticks to much older systems. AFAIK, NASA and others are still using the radiation-hardened RAD750 CPU for their most extreme applications (it was included in the Webb), which is made on either a 150 nm or 250 nm process—and costs ~$300k/unit!
They stick with it because they know it works, and its ~300 MIPS is sufficient for their rather modest local processing requirements. The large feature size is a benefit, since it reduces the ability of radiation to disrupt or damage the processor. I assume they have something similar for RAM.
RAD750 - Wikipedia
en.wikipedia.org
It’s also good to use III-V semiconductors with indirect bandgaps, because that way in order for a cosmic ray to flip a state it needs not only the right energy to excite an electron from the valence to conduction band, but it also needs to achieve the right change in angular momentum.
B01L
SlackMaster
- Joined
- Dec 1, 2021
- Posts
- 197
Potatoes, tomatoes...!
Seems to me the comparison to a PC workstation was a shifting of the goal posts since the Mac Pro only ever supported (up to, in certain models)128GB of RAM and made a fine OS X / macOS based workstation for years...
As it stands, with the 2019 outlier aside, the ASi Mac Pro supports more maximum RAM than any other Mac Pro to date...!
Yes, the comparison to a PC workstation is a shifting of the goalposts for the MP—but those goalposts were shifted by Apple itself! And we can hardly fault folks for assessing the MP in a way consistent with Apple's own shift in their presentation of the machine. Here's an excerpt from the 2019 press release...and they've done nothing since then, with their marketing of the ASi MP, to temper people's expectation that the latter should also offer workstation-class performance:
Apple today introduced the all-new Mac Pro, a completely redesigned, breakthrough workstation for pros who push the limits of what a Mac can do....Designed for maximum performance, expansion and configurability, the all-new Mac Pro features workstation-class Xeon processors up to 28 cores, a high-performance memory system with a massive 1.5TB capacity, eight PCIe expansion slots and a graphics architecture featuring the world’s most powerful graphics card.
Apple unveils powerful, all-new Mac Pro and groundbreaking Pro Display XDR
Apple today introduced the all-new Mac Pro, a completely redesigned, breakthrough workstation for pros, and unveiled Pro Display XDR.
www.apple.com
And it's not just the marketing, it's also the pricing that puts it into workstation territory. The starting price of the 2013 MP was $3,000, which was 1.5x the $2,000 starting price of the 2013 15" MBP.
By comparison, the starting price of the ASi MP is $7,000, which is 2.8x the $2,500 starting price of the 16" MBP.
Last edited:
