Article correction: The 20-core M1 Ultra is faster than the 28-core Xeon of the 2019 Mac Pro (not 28W). The W-3275M Xeon of the 2019 Mac Pro has a TDP of 205 watts.
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.
Alleged Apple Chip Plans Suggest 'A16' Will Stick With 5nm, 'M2' to Make Jump to 3nm Instead
- Thread starter MacRumors
- Start date
- Sort by reaction score
If anyone on twitter can make a prediction "leak" rumor, then let me make mine, you hear it first here, on mac rumors:
WWDC:
- New macs with M1X on 5nm slight stronger and more efficient than M1. Hey! M1X is tottaly not M1, so... M1 ultra is still the last M1 chip, now they're M1X, and that makes all the difference.
- New Imac Pro with an...(M,S,U,A are already used....) let's say E1 chip, a name exclusive for the mac Pro line of chips, 5nm, based on M1, with support for pci extensions and ram slot for upgrades + the unified memory on the SOC for faster memory.
Next Year:
- M2 using 3nm and ARM9
Source: My father Bill Gates from xbox who works at apple.
WWDC:
- New macs with M1X on 5nm slight stronger and more efficient than M1. Hey! M1X is tottaly not M1, so... M1 ultra is still the last M1 chip, now they're M1X, and that makes all the difference.
- New Imac Pro with an...(M,S,U,A are already used....) let's say E1 chip, a name exclusive for the mac Pro line of chips, 5nm, based on M1, with support for pci extensions and ram slot for upgrades + the unified memory on the SOC for faster memory.
Next Year:
- M2 using 3nm and ARM9
Source: My father Bill Gates from xbox who works at apple.
Maybe yours is hampered by company software. By countless users here and my own experience, the M1 is blitzing quick. Every app opens instantly.
This makes little sense from an economics perspective. Sticking with N5 when can move to N4 could save Apple money for the A-series super high production costs. N4 doesn't buy a whole lot for power savings but the logic density increase is 6% . N4 Also was always tracked as a 2022 high volume ( which means Spring 22 could go to high volume. N3 has been tagged 2H 2022 for a long time. That is simply too late for leading iPhone series since production has to start in April-Jume for that. That is Q2 not the second half. )
A chart here
TSMC Unveils N4X Node: Extreme High-Performance at High Voltages
N3 and N4P are 'late' for any target that needs to start in the first half of 2022.
A 6% logic desnity increase means Apple can get conservatively get 2-4% more dies out of a single wafers if they keep transistor budget about the same. Or get the same amount of dies out of a wafer if they grow the budget.
If they stick with N5P for another iteration and add more cores or transistor consumption, then they will get fewer dies out of a wafer than they did for the A15. The cost of dies would go up (wafer cost divided by fewer working dies means cost of die goes up). Given TSMC is raising prices on wafers that is even more money.
N3 is attractive because get a nice density increase, but its wafer costs are higher still. And would have been betting the farm on a fuzzy high volume start date several years ago. N4 would have been a safer path to lower costs pre-pandemic. It was an even safer path after the pandemic started to have major impacts on design tasks.
N4P doesn't really buy much in density ( more useful dies per wafer). It just lowers power consumption. Apple didn't have a big power consumption problem several years ago and really don't on the A15 either. ( If Apple did major funding for N4P maybe they get early tape out and high volume access so they would have taken that several years ago, but didn't necessarily need it badly. )
Now it is more believable that 3-4 years ago Apple though TSMC N3 would shipping in late 2022. ( High production could happen in early Q3 and wafers would be flowing out of pipeline by late Q4, so could do a December timeframe product. ) That appears to have gotten blown up. High volume is technically landing in the "second half 2022" timeframe so 'on time' but useful product of that high volume isn't going to pragmatically appear in useful amounts in 2022. Output from the pipeline is running a Quarter late. ( which goes to show why would have been a risky bad idea for the iPhone since Apple is trapped with that September dog and pony show and release. )
If TSMC can produce N45and N5P in equal volume at approximately the same yield rates , than the A16 is more expensive on N5P. ( a relatively smaller die incrementally increasing yield so even if N4 was slightly below N5P those could be offset. )
MediaTek is already shipping product on N4. If there was a "pipeline cleaner" product that needed to be run on the early high volume N4 runs to shake out the early bugs ... that has been done already. The A16 could just drop into that bow wave. [ TMSC doesn't have to depend upon Apple to pipeline clean their new processes. Other folks will do it. And small risk to Apple to let more folks do the 'cleaning'. ]
N3E largely isn't a 2023 product. N3E was suppose to be a year after N3.
"... Originally, TSMC planned to start high-volume manufacturing (HVM) using N3E about a year after N3 (i.e., in Q3 2023), but in the recent months a rumor emerged that TSMC was pulling in HVM of N3E by about a quarter due to better than expected test production runs. During its most recent conference call, TSMC confirmed that N3E's progress was ahead of schedule and that it was considering pulling in mass production using this technology, but did not elaborate about exact plans. ..."
TSMC Roadmap Update: N3E in 2024, N2 in 2026, Major Changes Incoming
N3E is likely in the early Q3 '23 range if N3 is in the late Q3- early Q4 range for '22. They might be able to do late June '23, but years ago that is not what the roadmap said. So any long term planning for the iPhone probably skipped it. June is late for an iPhone production start.
Pretty good chance Apple knew N3 would going to be a bit rocky transition so if it targeted late 22 then it would be relatively low risk to use that for a April-May start date for the A17. There would be a fair amount of slop time so that if it ran 1-1.5 quarters late then still would be starting up on time.
Similarly if Apple threw a relatively super duper low volume , but sky high price tag product at N3 for late '22 or early '23 then Apple could just take the lower than expected yield lumps and bury that in super fat "Utlra Duo" profit margins. ( > $1000 SoC as opposed to $100 Soc ). Apple could use the fat margins SoC to pipe clean the process or the small , high volume one on this iteration.
Apple doing A16 N4 , A17 N4P , A18 N3E would not be surprising. (A17 N3 an option). And put the Macs on a M2 N4 (plain , Pro, Max) , M3 N3 (across the board) , M4 N3E path ( plain , Pro , Max) path
Last edited:
Well, to be fair… as off-puting as the name/Twitter profile pic might be to some, ShrimpApplePro has been the most reliable Apple leaker as of late.
"The new Mac Pro, using our new P1 (Performance) chip." *And now everyone can without our consent, and while we turn a blind eye, call it the Power Mac once again.
In all seriousness...
If M1 Ultra was the last chip in that line, it doesn't mean it has to be the last with that similar architecture. They're just likely going to come up with some exclusion new moniker for the Mac Pro chip. Still very realistic that the chip actually looks conveniently like a further sandwiched together M1 Ultra x2 or 4.
In all seriousness...
If M1 Ultra was the last chip in that line, it doesn't mean it has to be the last with that similar architecture. They're just likely going to come up with some exclusion new moniker for the Mac Pro chip. Still very realistic that the chip actually looks conveniently like a further sandwiched together M1 Ultra x2 or 4.
…..And will be put out of business just like the Foxconn leakers under weird Instagram names that are all likely based out of China.
I am not 100% sure his wife and kids would agree but assuming they do, the name should be reserved for something very very very special. Maybe when they release the first 128 bit processor or something like that.
ShrimpApplePro has joined the chat. Welcome!
I never said anything about TSMC's 3nm process being delayed, just speculating that the "M1X" may have a limited release, if Shrimp is correct about the naming of it. (Which Shrimp has now said may actually be called "M2", but that wasn't stated when the article first ran.) What has been delayed are numerous Apple products because of China's zero-COVID policy, so Apple may not bother updating the entire product line with it.
It's not simply me speculating about it, Cliff Maier, the former Opteron architect who used to frequent these forums, says that his sources think that the M1X (or whatever they may call it) might have a limited run, with more substantial changes to the version using TSMC's 3nm.
That's not what Shrimp said in the original Tweet. Now they are saying that it may in fact be the M2, which suggests that Shrimp's alleged source isn't in marketing.
Sure, it could be based off the same M1 core (or perhaps an M1.5 core or even M2 core), but the point is that all the joke M1 xxxx names people have come up with in this thread don't make sense because as many other people have mentioned, Apple said the M1 line is complete.
BTW, some technical experts have claimed it would not make sense from an architectural point of view to build an M1 Ultra x 2 Mac Pro, whether that's two chips or one giant SoC, at least from their understanding of the architecture.
So, based on this, and based on the fact that Apple said M1 Ultra is the last M1 series chip, and also because of the probable requirement for vastly more RAM, I think the architecture is going to be somewhat different, hence the non-M naming I think it will get.
Last edited:
Hey, thanks for all that info dump, really interesting. Provided that you have a good, wider specter knowledge of the silicon details, I have a question: the GAA (gate all around) architecture won’t be ready from TSMC until the 2nm right? Any recent news on that? I expect that to come with a huge improvement in the instructions por clock, and thusly, to radically boost compute power or efficiency, but maybe I’m wrong, what do you think?This makes little sense from an economics perspective. Sticking with N5 when can move to N4 could save Apple money for the A-series super high production costs. N4 doesn't buy a whole lot for power savings but the logic density increase is 6% . N4 Also was always tracked as a 2022 high volume ( which means Spring 22 could go to high volume. N3 has been tagged 2H 2022 for a long time. That is simply too late for leading iPhone series since production has to start in April-Jume for that. That is Q2 not the second half. )
A chart here
TSMC Unveils N4X Node: Extreme High-Performance at High Voltages
www.anandtech.com
N3 and N4P are 'late' for any target that needs to start in the first half of 2022.
A 6% logic desnity increase means Apple can get conservatively get 2-4% more dies out of a single wafers if they keep transistor budget about the same. Or get the same amount of dies out of a wafer if they grow the budget.
If they stick with N5P for another iteration and add more cores or transistor consumption, then they will get fewer dies out of a wafer than they did for the A15. The cost of dies would go up (wafer cost divided by fewer working dies means cost of die goes up). Given TSMC is raising prices on wafers that is even more money.
N3 is attractive because get a nice density increase, but its wafer costs are higher still. And would have been betting the farm on a fuzzy high volume start date several years ago. N4 would have been a safer path to lower costs pre-pandemic. It was an even safer path after the pandemic started to have major impacts on design tasks.
N4P doesn't really buy much in density ( more useful dies per wafer). It just lowers power consumption. Apple didn't have a big power consumption problem several years ago and really don't on the A15 either. ( If Apple did major funding for N4P maybe they get early tape out and high volume access so they would have taken that several years ago, but didn't necessarily need it badly. )
Now it is more believable that 3-4 years ago Apple though TSMC N3 would shipping in late 2022. ( High production could happen in early Q3 and wafers would be flowing out of pipeline by late Q4, so could do a December timeframe product. ) That appears to have gotten blown up. High volume is technically landing in the "second half 2022" timeframe so 'on time' but useful product of that high volume isn't going to pragmatically appear in useful amounts in 2022. Output from the pipeline is running a Quarter late. ( which goes to show why would have been a risky bad idea for the iPhone since Apple is trapped with that September dog and pony show and release. )
If TSMC can produce N45and N5P in equal volume at approximately the same yield rates , than the A16 is more expensive on N5P. ( a relatively smaller die incrementally increasing yield so even if N4 was slightly below N5P those could be offset. )
MediaTek is already shipping product on N4. If there was a "pipeline cleaner" product that needed to be run on the early high volume N4 runs to shake out the early bugs ... that has been done already. The A16 could just drop into that bow wave. [ TMSC doesn't have to depend upon Apple to pipeline clean their new processes. Other folks will do it. And small risk to Apple to let more folks do the 'cleaning'. ]
N3E largely isn't a 2023 product. N3E was suppose to be a year after N3.
"... Originally, TSMC planned to start high-volume manufacturing (HVM) using N3E about a year after N3 (i.e., in Q3 2023), but in the recent months a rumor emerged that TSMC was pulling in HVM of N3E by about a quarter due to better than expected test production runs. During its most recent conference call, TSMC confirmed that N3E's progress was ahead of schedule and that it was considering pulling in mass production using this technology, but did not elaborate about exact plans. ..."
TSMC Roadmap Update: N3E in 2024, N2 in 2026, Major Changes Incoming
N3E is likely in the early Q3 '23 range if N3 is in the late Q3- early Q4 range for '22. They might be able to do late June '23, but years ago that is not what the roadmap said. So any long term planning for the iPhone probably skipped it. June is late for an iPhone production start.
Pretty good chance Apple knew N3 would going to be a bit rocky transition so if it targeted late 22 then it would be relatively low risk to use that for a April-May start date for the A17. There would be a fair amount of slop time so that if it ran 1-1.5 quarters late then still would be starting up on time.
Similarly if Apple threw a relatively super duper low volume , but sky high price tag product at N3 for late '22 or early '23 then Apple could just take the lower than expected yield lumps and bury that in super fat "Utlra Duo" profit margins. ( > $1000 SoC as opposed to $100 Soc ). Apple could use the fat margins SoC to pipe clean the process or the small , high volume one on this iteration.
Apple doing A16 N4 , A17 N4P , A18 N3E would not be surprising. (A17 N3 an option). And put the Macs on a M2 N4 (plain , Pro, Max) , M3 N3 (across the board) , M4 N3E path ( plain , Pro , Max) path
Anyways, I don’t think I’ll wait until 2nm to make the jump to Apple Silicon, and will probably buy the Mac mini or MacBook with the M2 or M3.
With supply shortages and costs going up. I think Apple will find ways of giving us the same and say its more. And then take the chips that should have replaced the current ones and calling those pro and charging way more for those. The same old Apple but even more aggressive on higher profit margins than usual. Until prices go down and they can give us a $100 discount on computers we are going to be paying several hundred more now because of the supply constraints.
clearly someone who doesn’t know how CPU development works.
it’s easier to mass-produce a new process/cores on smaller less demanding versions, then scale up as the new process is standardized.
This goes for pretty much all manufacturers, Apple, Intel, AMD.
Apple has been doing this for years.
Just look at the “A” series of processors, every new generation is always introduced in the most basic edition first, and then the “A-X/Z” come later, if they come at all as Apple has completely skipped generations of A-X before.
There was no A7X, no A11X, no A13X.
Same with the Mac, it’s been happening for years.
As an example take the 2012 IvyBridge processors.
First came to The MacBook Air and MacBook Pro in June 2012, then the Mac mini and iMac in October 2012, then the Macpro in… Late 2013, after most of the lineup had already transitioned to the next generation Haswell.
That’s just how it works.
Love posts like this.
The last Mac Studio wasn’t even released a full three months ago, it’s still back ordered months, the next one most likely won’t be here for another two years at the least.
Once that one comes out they’ll be another one around the corner two years later, why not wait for that one?
Majin Bu has leaked on Twitter claimed diagrams of Apple's UltraFusion Chip Interconnect in both horizontal and vertical orientations and this would allow Apple to connect two M1 Ultra SoCs together along their vertical edges.
So essentially you would have four M1 Max in a square configuration connected on two sides each using the UltraFusion Chip Interconnect.
Yes it makes economic sense for Apple to use N4 or N4P. But this report says Apple may instead stay on N5P.
If they do stay on N5P, the most-likely reason that comes to mind is that Apple is not willing to forgo millions (if not tens of millions) of iPhone 14 Pro family sales if TSMC's supply falls short over both the launch quarter and follow-on quarters.
Back in early 2021, the general speculation was iPhone 14 (A16) would be on N3, but as TSMC encountered issues ramping it up, by late 2021 the general speculation had changed to iPhone 14 (A16) would be on the N4P and M2 would be on N3 (and launch with the iPad Pro refresh).
Because it had "N4" in the name, everyone at first assumed N4P was 4nm (like N4), but TSMC later clarified that N4P was 5nm, but with sufficient enhancements that lowered power consumption by 22% and increased circuit density by 6% (as compared to the current N5P used in A15 and M1, which offers 10% lower power consumption or 5% more power than N5 at the same transistor count).
It's been generally accepted within the industry media that N3 would not be ready at sufficient scale to support A16 and therefore it would be used for A17 (and M2).
As N4P is a better N5P, Apple might have been planning to use it for the A16 (since those are only going into the iPhone Pro family) and this latest rumor implies that not enough might be available to meet Apple's needs and therefore A16 will also use N5P (as the A15).
But do we have a firm idea of how many N4P SoCs TSMC can make? Apple will need tens of millions of them to support the iPhone 14 launch through Q1. And then scores of millions through the remainder of the sales cycle until Q1 2023 and the iPhone 15.
If TSMC cannot provide that volume, then Apple has little choice but to fall back on N5P unless they just want to lose (tens of) millions of iPhone Pro sales due to lack of supply.
Last edited:
TSMC has an N4X - same as N4P except that it can have the voltage increased up to 10x the N4P.
Oh, and it’s been shipping since mid-December 2021 to all customers.
Oh, and it’s been shipping since mid-December 2021 to all customers.
Agreed. like someone else stated on page 1 of this thread seems chip fabrication delays or final product assemblies are affecting order fulfillment it’s likely an adjustment was done to speed up M2 chip timeline, along with Microsoft’s Dev Week announcement of ARM based CPU’s having CPU+GPU+NPU as well.
All I can think about this M2 chip is Tubbs from Miami Vice TV Show back in the late 80’s saying ‘You Beast You’ with a wide ear to ear grin (he stole Crocket’s speed boat).