Now that we have the M3 series out I think we can probably begin to predict when Apple will release its next chips.
....
There is still a decent amount of noise in trying to predict this because some very relevant factors were out of Apple's control over the last 2-3 years when M1-M3 were being developed/prepped/etc.
Pandemic had substantive logistical impacts. TSMC's N3 rolled out and ramped relatively slower than previous generations. Both of those had some impacts on shipping times for Apple products.
The M3(3nm) series is unique in that pretty much all of the chip family other than the Ultra and a possible “extreme“ version was introduced. Apple also apparently has a deal with TSMC so to make sure they had all of the 3nm production and a jump start on 3nm tech, I would guess this is why they released all these machines when they did, and all next year we will see iPad and lower end Mac refreshes throughout the year using the M3.
The "all of 3nm production" meme is grossly overblown. There is a huge difference between Apple bought it all (to block everyone else ) and nobody else bought it ( because of bad timing and/or too expensive). TSMC when into high volume production on N3 back in December 2022. If Apple is only customer buying anything for the 1H 2023 then it wouldn't be had to stockpile some M3's to put into a relatively lower volume product. The iMac isn't the super dominate desktop anymore in sales. The MBP 14"/16" prices keep them from being mega volume sellers also.
MediaTeK shipped a TSMC N4 SoC before Apple did. There is large myth that Apple always gets everything first. Before Huawei got blocked they too were getting access about as early. There were rumors that Intel had bought up all the early N3 . And then Intel slid on deliver time ... shocker ... and they moved back in the queue for a later roll out. Intel bumbling and Apple buying the addition wafers that surfaced in availability probably means they didn't buy them all. It is just that no other player with a large enough demand footprint to matter showed up.
Current leaks point to Intel's N3B stuff coming to OEMs in mid 2024.... about 6+ months later than their roadmaps from several years ago.
Intel stumbling here probably is not a firm foundation to make future timeliness predictions on. Some of this is Intel flip-flopping on focus and design. Some folks wanted to exit foundry business and others didn't. The N3B cores Intel is going to roll out will also be implemented on Intel 20A. Intel is going to load up on TSMC and then reverse itself over the next 2 years on the CPU tile/chiplet front.
Because of the Apple Vision Pro, I can’t see Apple releasing any of these new computers until WWDC, with the last of them coming around Oct of 2024.
So far the Vision Pro is still on M2 . So how is the VPro consumption of M2 SoC packages going to disrupt the M3 roll out? It is two different dies made in two different factories. Even more so if look at the M2 Pro and M2 Max die production that is likely dropping pretty fast. Apple stopped selling M2 variants of the MBP. So where are all those wafers going? Not away from the VPro. Apple transitioning Mac products off of M2 actually increases M2 availability to the VPro. Therefore it also increases being able to produce the two separate lines concurrently. The exact
opposite of what you are hand waving at here.
Pretty good chance the Mini ( M3 , M3 Pro) would get an update substantially before June 2024. Getting to Apple's highest volume product ( MBA line) could take until June , but the relatively lower volume ones are easiest to do. ( The Air's 'problem' is the MBA 15" which is still relatively new ).
The Ultra Mac Studio and Mac Pro also have 'problems' with MBP 16" Max's competing very close on some benchmarks. Nevermind the Zen5 (and incrementally fixed Intel workstation offerings) and AMD/Nvidia new GPUs queuing up for 2024. It would not be surprising of Mac Studio/Pro snoozed until June 2024. Pretty good chance Apple expects the M3 Max to sell well enough in laptops to sell all the inventory of that SoC they have queued up to make over the next 3-4 months.
Even if Apple did a last minute switch of the VPro from M2 to M3 before its initial launch (or if the R1 is on TSMC N3 ), that wouldn't impact roll out cadence at all in terms of 'diverted volume'. Projections are that Apple is only going to sell a several 100K VPro systems. The MBA selling in double digit millions. It is a different order of magnitude in volume. Selling a couple 100K less MBA is a 'normal' quirky month of sales that averages back out the other way another quirky upside month later.
The real substantive demand bubble Apple would need to time shift for would be for iPhone demand on N3B wafers. Otherwise, just incrementally updates the Macs through the 1H of the year to distribute the load. The deeper get into 2024 the more other N3 customers there are going to be. Apple trying to heard all their M-series products into 1-2 mega big bang roll outs with a dog-and-pony show per year is goofy.
If an ”extreme” were to be released, I wouldn’t expect it until late 2024, early 2025. All in all, I would guess this processor family goes for 16-18 months before M4 is released in 2025.
With the GPU improvements Apple threw at the M3 Max creeping into the M2 Ultra range on some tasks, I doubt there is an "extreme'. Two M3 Maxes could be in the range of 4 M2 Maxes , which gets them the 'extreme' without all the Rube Goldberg complexity of trying to mash a monolithic chip into a quad 'chiplet' .
the 'extreme' (relative to M1/M2) likely gets done by June-August as a M3 Ultra.
The M4 (3nm improved manufacturing process)
What successor isn't an improved manufactuing process? N7 -> N6 improved. N5 -> N4 improved. N3E -> N3P improved.
N5 family -> N3B is improved on density. It isn't as much of an 'improvement' on costs though.
Folks keep waving at N3E was though it is an across the board improvement over N3B , but it backslides on density on logic. It backslides all the way back to N5 on cache/SRAM.
if the M2 is like a “M1S” then the M4 would definitely be the “M2S” I think a lot of this really depends on if they have any new products that could take advantage of the 3nm chip such as an updated Vision Pro, but judging by how the M2 generation was released, we will get these chips around WWDC 2025,
WWDC timing is like a broken analog clock. It just happens to be correct twice a day. There is about zero good reason for Apple to anchor Mac around WWDC . For a while Mac line ups were dominated by x86 laptops. The Computex convention would typically happen a week or so before WWDC each year. When Intel was focused on making a fuss about mobile CPUs each year it make lots of sense for them to target that. When Apple was on more of Intel cadence ... that just fell out of the broader PC market cadence.
Since the M-series and A-series introductions often compete for the same wafer capacity, it is going to make far more sense for Apple to move at least a decent fraction of Mac introductions away from the 'stuck in time' iPhone demand bubbles. ( at some point those stuck in time iPhone intros are going to degrade the iPhone also. It is still mostly working OK . ) . WWDC is a bad timing if trying to do some deliberate decoupling.
with the last of them probably released around October/November of 2025. I would guess this generation last even shorter Than the M2 generation did, at around a year tops as Apple likely is expecting demand to soften with these chips as well.
softening demand for the larger dies will likely lengthen their time on market rather than shorten it. The higher development costs makes an increasingly shorter product life cycle even more problematical. The Mn Max doesn't really have a 'hand me down' product to be shoveled into. The iPhone A-series diverts to AppleTV , iPad , etc.
Apple can keep making 'new' products for multiple years.
Killing off a Mn Max (or larger ) every 12 months is a problem. [ Even if AMD/Intel intro new server chips every year they don't stop selling the old ones right away. There is still old products that go out the door with the older chips. The dual edge sword for the Apple Silicon group is that they only have one and only one customer. Once a big die mac SoC gets dumped by the large , low volume Mac product there is no where else to go. One customer with 1-2 products which both dump around the same time. ]
The M5 (“2nm” my understanding its not really 2NM just a far more dense 3NM)
None of these are directly measured 'nm' so they are
all in the category on not really being ' xx nm '. TSMC has more dense N3P and N3S coming after N3E.
The whole 'nm' suffix is just pure marketing hookem. Just drop it. Intel , TSMC , Samsung all have stuff like "Intel 3" , N3P , S3LPP to use. Sentences with 'real' quickly followed by 'nm' are basically an oxymoron.
The numbers are no basically a number to indicator multiple dimension density. It is not a unidimensional , linear measurement. They are all synthesized and more so inductive of generation progress than direct measurement.
