[Rumour] M2 chip exclusive to Mac Pro and iMac coming in the 2nd half of 2021

[deconstruct60]

I believe the number will correspond to the new chip fabrication process just how they do it with the A-series.

M1 = 5nm process (2020)
M2 = 5nm+ process (2021)
M3 = 3nm process (2022)
M4 = 3nm+ process (2023)

So all the Mac chips coming out until the new 5nm+ process later next year will all use the M1 name with additional letters to denote additional core count and other features. So we’ll likely see something more like this:

If Apple does it "Just lie the A-series" this is exactly how it will not look.

There is no A11X , nor A13X. The letter modifier iPad Pro SoC A--X versions skipped intermediary fab process nodes to do bigger transistor budget bumps with major fabrication node shrinkage. The letter modifiers skipped "+" nodes and/or design re-optimization on same node.

Once the full Mac line-up is on M-series processors Apple is also unlikely to want to put the M-series into fab capacity contention with the bleeding edge A-series. Even if smaller in volume, chip dies that are 2x-3x large than the iPhone dies have multiplier in wafer start consumption. As Apple's dies get bigger their ability to 'race' other chip implementers to the most expensive fabrication process avaialble will go down. ( Especially when lots folks are scrambling over the exact same set of production machines. )

M1 = MBA, MBP 13” 2-port, Mm 2-port
8-core CPU

M1X = MBP 13” 4-port, Mm 4-port, MBP 16”, iMac 24”
12-core CPU

M1Z = iMac 30”, Mac Pro mini
18-core CPU

Doubtful that M1-series gets past 18 cores on 5nm. And more likely just an addition 4 cores like the jump from 8-> 12. So the M1Z with a iMac 27" , iMac 30-32" , 'half" Mac Pro (mini). The SoC stretched over multiple form factors.

Pretty good chance that Apple calls the Mac transition "complete" with the "Half sized" Mac Pro 'mini'. They go back into Rip van Winkle mode again (well actually probably already are in slumber mode on the primary system/chassis. ) on the full size Mac Pro and come out with something later in the M2 line up when they have more access to a 3nm shrink. I doubt Apple really wants to do that chip. Pretty likely past the two year deadline they gave themselves. But even then I'd dobutt they'd shoot for 32 cores. Maybe something in the 20-24 range. They probably are not going into the "core count" wars that the Threadripper, EPYC , Xeon SP , and datacenter ARM Neoverse N1 ( N2 , etc.) implementers are going to do.

Apple will add that much larger I/O needed for the larger case with far more slots, but "faster than the old Intel Mac Pro " will probably be in the sufficient enough requirements for them.

 

[ADGrant]

Apple M1:
MacBook Air
MacBook
Mac mini

Apple M2:
14" MacBook Pro
24" iMac
30" iMac

Apple M3:
16" MacBook Pro
Mac Pro mini
Mac Pro
?

I don't see the 16" MBP waiting for a third generation of Apple Silicon. I also don't think the desktops Macs will have the same SoCs as the MacBooks (the current Mac Mini will be the exception IMHO). I would not be surprised to see the Desktop SoCs with fewer or no efficiency cores.

 

[deconstruct60]

I don't see the 16" MBP waiting for a third generation of Apple Silicon. I also don't think the desktops Macs will have the same SoCs as the MacBooks (the current Mac Mini will be the exception IMHO). I would not be surprised to see the Desktop SoCs with fewer or no efficiency cores.

the downclocked GPUs of the iMac 21.5" are not too far from the downclocked GPUs of the MBP 16". If Apple had something to cover the MBP 16" GPU wise that could probably cover the iMac 21-24" also. Apple has never made the "smaller screen" iMacs GPU "speed daemons. And that is one less SoC derivative they have to create.

They could throw that M1X into the upper range of the Mini. iMac 21-24" , MBP 13" four port , and MBP 16".

One of Apple's apparent objectives here is to remove dGPUs from the line up as much as they can. The "mobile" (downclocked) lower 40% range GPUs in the
I don't think they are going to try to remove dGPUs from all of the systems. Just the relatively higher volume ones ( lower prices, higher volume iMacs and all the laptops ). [ the desktop Mini never had one except for corner case flukes (or no iGPU available at all back in PPC era. ) ]

The Mini getting a "laptop" processor is probably not an exception. If go back to WWDC presentation, one of Apple's primary objectives is to move desktop's to better perf/watt. iGPUs do that. Keep the Mini and most of iMacs cool has been an issue (when coupled to Apple's case design requirements.). They probably are going to use the M1-series to "ease" the thermal pressures there.


The higher clocked , much bigger die GPUs I don't think they are going to try to cover. But at that point in the upper end of the iMac 27" ( which isn't most of the iMacs sold) , iMac Pro , and Mac Pro range.

All of that will even further solidify Apple's GPU as being the primary target and focus of macOS developers. That is an even better hammer to push conversion to Metal for Mac apps and into specific optimization code in Metal for Apple's GPU.


There is a pretty good chance that the upcoming A14X is just the same die as the M1 with some functions turned off/on. [ The Mini's limitation of just one display out the TB ports is just one of several indicators; including the size of the die. ]. Apple doing something "super exclusionary" special just for Macs. That probably didn't surface in the M1. Probably something incrementally better in the M1X ( a bigger version. ). But Apple's presentation indicated they are about more focused on pushing laptops into the classic "desktop" performance pace than on focusing monomanically on desktops ( given sell more than 70+% of laptops ... that's lined up with the users and money also. ).


The major piece that Apple is missing from their iGPU is more local fast cache ( something like AMD "Infinity Cache" , some eDRAM , and/or perhaps 2GB HBM2 backside cache). Crank up the number of dispaly output engines and much larger frame buffer cache (that won't thrash the system level cache) and they are probably done with several dGPUs in the Mac line up.

Some relatively small delta to get to 4 USB Type-A ports and one 10GbE port (with discrete controller) they are probably done for lower 'half' of desktop line up.

 

[leman]

1. Apple hasn't had the bigger iPad Pro SoC ( A10X , A12X , . probable A14X ) on a 12 month cadence for a long while. Apple has typically waited for process shrinks to bump their larger dies.

What you say makes a lot of sense. What I meant is that an "M2" (whatever it is) won't be based on A14. It could be A15 or A16 (if your assumption is correct and Apple will adopt a slower release cadence for Mac hardware)

[ The phones don't particularly need SVE2 and "other pro features" ( large RAM capacities , mid double digit PCI-e lane provisioning, etc. ) that would bloat out the die size for the phone. Apple has their own AMX matrix like extensions and can probably tune up SVE on 5nm or 5nm+ for the phones on the A15 iteration. ]

This is true, but at the same time the higher-end Pro Macs could really benefit from having SVE. And an argument can be made that hardware feature fragmentation is not good for the ecosystem. It seams to me that Apple is moving towards feature parity across the platforms: same basic functional capabilities, different performance levels and limits.

The question is also how expensive such a feature would be. Maybe it is actually cheaper for Apple (in terms of hardware design) to implement SVE across the board than to only introduce it on a selected few chips. For example, iPhones don't need four 128-bit vector units either, but they do have them.

The desktops are more likely to get some A14 derivation on a larger die.

You make a convincing argument. This would of course mean no SVE on first iteration of higher-end Macs, which might be a missed opportunity for Apple. But then again they could introduce it in 2022 and say "hey, look, now numerical computation is 30-40% faster, time to upgrade!"

 

[leman]

I would not be surprised to see the Desktop SoCs with fewer or no efficiency cores.

Efficiency cores make a lot of sense on desktop too — offloading backgrounds tasks onto slower energy efficient cores leaves more resources to more interesting tasks. Being able to do backups etc. without significant impact on system performance is a valuable ability. I think we'll see 4 efficiency cores across the board, don't need more than that and they are not taking much space on the die anyway. But who knows what Apple decides.

 

[ADGrant]

Efficiency cores make a lot of sense on desktop too — offloading backgrounds tasks onto slower energy efficient cores leaves more resources to more interesting tasks. Being able to do backups etc. without significant impact on system performance is a valuable ability. I think we'll see 4 efficiency cores across the board, don't need more than that and they are not taking much space on the die anyway. But who knows what Apple decides.

The purpose of efficiency cores is to lower the power consumption required to run those tasks (and allow the SoC to run cooler). For a machine with a good cooling system plugged into the wall, those constraints are not as important. There are other things that could be put on the die such as more full power cores or GPU cores.

 

[leman]

The purpose of efficiency cores is to lower the power consumption required to run those tasks (and allow the SoC to run cooler). For a machine with a good cooling system plugged into the wall, those constraints are not as important. There are other things that could be put on the die such as more full power cores or GPU cores.

Look at the die shots of M1 - the entire icestorm cluster with four cores take approximately the same area as a single firestorm core. I have a strong suspicion that it’s more than just power conservation. It’s likely better to have 8 high performance cores free do do work while lower priority tasks run in the background than having 9 high performance cores that have to do the main work and the background tasks.

 

[vladi]

The ARM64 Mac Pro will most likely have stackable processor cards.

I.e. the upgrade path is adding another 16 core “card” with XGb of memory or whatever.

The cross connect/cache coherency between “cards” will be very interesting.

Apple solders everything on PCB and that's been the trend in their hardware for quite some time. I highly doubt you will be getting option of SoC upgrade PCIe or something stacks. At this point I believe small MacPro will be only customizable on a website build and that's it.

 

[JWD]

I'm in for the M1S ?

 

[richinaus]

I would like to see then double the Fast CPU cores and GPU cores. That should enable them to beat the current crop of processors for all other products minus the Mac Pro. For GPU especially, if the M1 GPU is ~5500xt performance, then double that should be ~5700xt performance. Given that M1 performs at the level of the desktop cards, then the graphics at the high end should also still see an increase in performance relative to the current crop of Discrete GPUs in their other products. It also means less differentiation, at least currently, with better GPUs at the current price points, so that the high end GPU upgrade no longer exists-you get the upgrade for free essentially. Add in double of the TB lanes, etc. and they will have a product with features that match or exceed most other products. In the case of the Mac Pro, they would have to triple or more the CPU cores to get even close to the performance of the high end processor, but that’s a halo product, so probably the last to get updated...

I am very intrigued to see Apple is going to get close to the latest high end desktop GPU’s.
The cpu is sort of expected now but the GPU is a different ball game especially in pro desktops.
Plus the apps need to support it. The ones I use mostly aren’t even ready for M1 chips......

 

[senttoschool]

Apple M1:
MacBook Air
MacBook
Mac mini

Apple M2:
14" MacBook Pro
24" iMac
30" iMac

Apple M3:
16" MacBook Pro
Mac Pro mini
Mac Pro
?

No, this is not how it will work.

Apple releases one new SoC design each year. This same SoC will scale from iPhones to iPads to Macbook Airs to Mac Pros. It will scale by having a different number of cores and/or different wattages.

The Mac Pro might get the M2 because, by the time it releases, Apple has already transitioned away from the M1 core. It has to do with timing. It's not because Apple wants to differentiate by using different architectures for different machines since Apple Silicon can scale well between low power to high power devices.

 

[EntropyQ3]

I think two things are missing from these technical discussions.
One is - "What are the driving use cases?"
The other is - "What is the likely market volume?"

Now, these are to some extent connected, and a bit nebulous. After all, computers are general purpose devices, and there are a lot of smaller niches that could conceivably add up to respectable volume. And corporate pride/halo products can be a factor too.

But just doing technical speculation without having any kind of relevance analysis is a bit of a dead end. Apple can do pretty much anything, the question is what makes sense for them to do. And this mostly affects the Mac Pro. Personally, my crystal ball gets awfully foggy once we go beyond, say, 16 performance cores and 32 GPU cores, because I just can't see substantial market volume beyond that, but that isn't a guarantee that Apple won't do it for a variety of reasons. I simply haven't seen anyone provide those reasons beyond it being a technical possibility.

(Although, in a sense, just giving your mind access to 3nm lithography and freedom from economic concerns sure is liberating. Maybe that is what Apple has done just to see if their engineers can come up with something new and compelling. But if so, I doubt it would be "twice as much of everything" or "a really powerful x86 PC only with our silicon".)

 

[leman]

But just doing technical speculation without having any kind of relevance analysis is a bit of a dead end. Apple can do pretty much anything, the question is what makes sense for them to do. And this mostly affects the Mac Pro. Personally, my crystal ball gets awfully foggy once we go beyond, say, 16 performance cores and 32 GPU cores, because I just can't see substantial market volume beyond that, but that isn't a guarantee that Apple won't do it for a variety of reasons. I simply haven't seen anyone provide those reasons beyond it being a technical possibility.

I think one of the factors is “prestige”. Mac Pro is a niche machine, but it is a symbolic product. By having the very expensive and very powerful Mac Pro in their lineup, Apple signals that they care about professional users and Mac users in general - even if very few of them will every buy one. I think it makes sense for Apple to have a Mac Pro even if the product itself is not profitable.

 

[UBS28]

I say it is ********. Apple does not need to wait until Q3 2021 for adding simply more cores for the 16” MBP version. If it will simply be the M1X in the 16” MBP, they could have launched it in Q1 2021 already.

The M2 / M2X is much more likely for the 16” MBP than a simple M1X chip.

 

[winna]

Rumour, M23 coming to Mac 2057

 

[EntropyQ3]

I think one of the factors is “prestige”. Mac Pro is a niche machine, but it is a symbolic product. By having the very expensive and very powerful Mac Pro in their lineup, Apple signals that they care about professional users and Mac users in general - even if very few of them will every buy one. I think it makes sense for Apple to have a Mac Pro even if the product itself is not profitable.

Prestige/halo/corporate pride is definitely something that would justify speculating about an all out performance effort.
If we use that as a base assumption however, the article and thread title makes no sense, since a 64-core CPU coupled to a 128-core GPU hitched to a 4096-bit wide HBM2e memory system (for example) doesn't really fit the iMac which should have a reasonably consumer friendly price tag, monitor included.

If we limit ourselves to configurations that would fit the large iMac as well as the Pro, then we are suddenly constrained by power (particularly if the new enclosure is even more narrow) and to some extent cost.

Economically it may well make sense for Apple to spread the SoC design cost over the large iMac as well as the Mac Pro, but could the performance achieve much in the way of halo then? The AMD Threadripper Pro systems sold by Lenovo with up to 64 cores and with 8x64-bit memory, coupled with an Nvidia 3090 or the big Quadros put the bar pretty high. And you can buy that now, not in a year.

 

[leman]

Prestige/halo/corporate pride is definitely something that would justify speculating about an all out performance effort.
If we use that as a base assumption however, the article and thread title makes no sense, since a 64-core CPU coupled to a 128-core GPU hitched to a 4096-bit wide HBM2e memory system (for example) doesn't really fit the iMac which should have a reasonably consumer friendly price tag, monitor included.

I agree. Making large chips Mac Pro exclusive is going to be extremely expensive and unlikely to be profitable at all, but there are other solutions like chiplets (in this case Mac Pro would "only" need a custom I/O die and memory packaging). Then again, who knows. It's Apple we are talking about. After all, 3090 die is 628mm2 large and I doubt that Nvidia sells too many of those GPUs — must still be worth it somehow... Maybe Apple will just build a 1000mm2 large monolithic die for the Mac Pro that costs $6000 to manufacture per piece and then sell it in a $15000 tower...

 

[vladi]

Biggest bottle neck for M chips is Big Sur. It's riddled with little bugs and stuff on M1 but Apple is so slow at delivering the updates. That is my biggest concern. I haven't turned on my M1 Mini in weeks. Featherweight apps are coming up with ARM versions while heavyweight apps are no where to be seen but they will come out eventually I'm not worried. I'm more worried about Apple's OS development. This is going to be at least a three year transition and if they release a tower this year it's only going to be for marketing purposes like to be co-released with Adobe CC apps or whatever.

 

[grjj]

No I mean 12. I am already running 11.1

It took a decade for macOS to go from 10 to 11. You really think it will go from 11 to 12 in a year?
Not saying a hard "no", it just seems REALLY unlikely.

Remember that the CPUs changed from Pentium to Core, the name changed from Mac OS to macOS, the entire underlying framework changed (Carbon/CoCoa). Point is that Apple doesn't change the major version lightly any more. I'm not saying they can't, but for all its "What's next?" attitude, there are parts of Apple's though process that are steeped in tradition. A bump in CPU performance has never been the driver of a major revision increase.

 

[ArPe]

It took a decade for macOS to go from 10 to 11. You really think it will go from 11 to 12 in a year?
Not saying a hard "no", it just seems REALLY unlikely.

macOS Big Sur uses iOS style numbering. Thought this was obvious by now.

 

[dmccloud]

If Apple does it "Just lie the A-series" this is exactly how it will not look.

There is no A11X , nor A13X. The letter modifier iPad Pro SoC A--X versions skipped intermediary fab process nodes to do bigger transistor budget bumps with major fabrication node shrinkage. The letter modifiers skipped "+" nodes and/or design re-optimization on same node.

Once the full Mac line-up is on M-series processors Apple is also unlikely to want to put the M-series into fab capacity contention with the bleeding edge A-series. Even if smaller in volume, chip dies that are 2x-3x large than the iPhone dies have multiplier in wafer start consumption. As Apple's dies get bigger their ability to 'race' other chip implementers to the most expensive fabrication process avaialble will go down. ( Especially when lots folks are scrambling over the exact same set of production machines. )

Apple has already booked out a full year of 5nm production from TSMC. Once they move down to 4nm or 3nm, that will free up AMD to move their Ryzen parts from 7nm to 5nm. So they are are actually getting those machines all to themselves at the moment, leaving 7nm and larger processes as a free-for all for everyone else. TSMC has more than enough production lines and processes running to supply all of its clients (although this semiconductor shortage will affect everyone equally).

 

[dmccloud]

It took a decade for macOS to go from 10 to 11. You really think it will go from 11 to 12 in a year?
Not saying a hard "no", it just seems REALLY unlikely.

Remember that the CPUs changed from Pentium to Core, the name changed from Mac OS to macOS, the entire underlying framework changed (Carbon/CoCoa). Point is that Apple doesn't change the major version lightly any more. I'm not saying they can't, but for all its "What's next?" attitude, there are parts of Apple's though process that are steeped in tradition. A bump in CPU performance has never been the driver of a major revision increase.

Mac OS may not jump from 11 to 12 in one year, but it will jump within 2 years. With previous version numbering, updates to the OS were in the 10.xx.xx format, with the second number being the OS version and the third number being the update version. With Big Sur, Apple switched to 11.xx for version numbering. If Apple was going to stretch out Mac OS 11 as long as they kept the OSX/Mac OS 10 naming conventions, we would be on 11.1.2 right now instead of 11.2.

 

[Lemon Olive]

Apple loves to say "there's never been a better time to buy a Mac". Well, if you're a Pro or even Pro-sumer there has never been a worse time than right now. With already-announced transition, and the first run of low end Apple Silicon Macs already out...the wait now for an iMac, MacBook Pro, or any machine with an Apple Silicon chip more capable than the M1 is painful.

The M1 Mac Mini is twice as powerful as my work iMac, but has half the amount of memory, and leaves me without an Apple quality display.

 

[pshufd]

Apple loves to say "there's never been a better time to buy a Mac". Well, if you're a Pro or even Pro-sumer there has never been a worse time than right now. With already-announced transition, and the first run of low end Apple Silicon Macs already out...the wait now for an iMac, MacBook Pro, or any machine with an Apple Silicon chip more capable than the M1 is painful.

The M1 Mac Mini is twice as powerful as my work iMac, but has half the amount of memory, and leaves me without an Apple quality display.

You still have all of your options - but the fact that better systems are coming is a downer for what you want to do now. So you make do with what you have or buy something as a stopgap if you need to. Or you buy some used hardware to tide you over. The new stuff coming out may do a number on the used Intel market.

 

[Tagbert]

Apple M1:
MacBook Air
MacBook
Mac mini

Apple M2:
14" MacBook Pro
24" iMac
30" iMac

Apple M3:
16" MacBook Pro
Mac Pro mini
Mac Pro
?

I don’t think that the naming makes sense like that.

For the iPhone chips, they have the family (A) followed by the generation (14) and then sometimes a performance edition (X/Z). This allows them to increment the generation and use the suffix to indicate performance levels.

I could see them doing something like this

M1
Air, Mini, low end MBP

M1t - more RAM, more perf cores, more GPU cores, more ports
MBP, low end iMac

then when then A15 chip is ready, they can do an M version

M2
Air, Mini, low end MBP
(may not ship until 2022)

M2t
MBP, low end iMac

M2s - even more memory, even more cores, discrete GPUs?
Mac Pro, high end iMac