Next-Gen Apple Chip Technology on Track to Reach Production in 2025

[anakin44011]

Still loving my M1 Max MBP - she's a rocket. I used to switch out Macs every year or two, but I think I'm going to keep this one for 3-4 more years, its so good at what it does.

I hear you...

Benchmarks are fun to look at, but many of the real-world performance increases are software dependent. So until the software that you use will enable you to "experience" the difference in an M2 series, M3 series, M4 series, etc, you'll be fine with your M1 Max.

And that, fortunately or unfortunately, is largely out of Apple's hands.

 

[deconstruct60]

​

TSMC chip fabrication facilities will begin installing equipment designed for 2nm chip production in April at the earliest. Apple was the first company to utilize TSMC's initial 3nm technology with the A17 Pro chip in the iPhone 15 Pro and iPhone 15 Pro Max, and the company is likely to follow suit with the chipmaker's 2nm chips. With production set to begin in 2025, 2nm chips are likely to make their first appearance in Apple devices soon after.

This is similar 'hype train' that got whipped up for the TSMC N3 process. N3 went into high volume production in Q4 2022. Apple's N3 chips didn't come until Q3 2023. Almost a year later.

N2 is likely coming 2H 2024. N3E when high production 2H 23 ( and no Apple chips showed up for that also). TSMC has a N3P and N3X coming also in 2024. The paragraph above says that the equipment hasn't even been installed yet ( and it has its own post install validation ramp to go through also. )

Other coverage in Anandtech a while back after TSMC gave a status update earlier.

" N2 ... H2 2025

...
N2P 2026 ..."

TSMC Outlines 2nm Plans: N2P Brings Backside Power Delivery in 2026, N2X Added To Roadmap

www.anandtech.com

More than decent chance N2P will also trail N2 by about 12 months; so yet another H2 2026 landing.


The problem with these re-occuring H2 202x dates is that they are all pragmatic misses for the A-series SoCs. The iPhone launches are stuck on a entirely non technical dates in September. These chip process rollout outs are not. In order to have sufficient volume Apple needs finished chips in April-June (Q2 ) not late Q3. Pretty likely N2 has an even longer 'bake time' than N3 does (which is already several months). [even more multipatterning on current tech EUV machines] .


The R-series or M-series rollout out are much more likely matches to the timing than the A-series.


On TSMC N4 Mediatek rollout out before Apple did. The A-series being boat anchored to huge demand bubble launches in September means from time to time Apple isn't going to be the first one out of the gate on TSMC Process. From time to time some other vendor is going to have a more flexible schedule and get out the door earlier.

Last month, the Financial Times reported that TSMC had already demonstrated prototype 2nm chips to Apple ahead of their expected introduction in 2025.

TSMC has prototype N3 chips too in 2021-22 timeframe and those didn't ship until H2 2023. They aren't done with the work to get to volume production.


N2 isn't just 'smaller'. It is a shift to "gate all around" / "nanosheet" technology that TSMC has not done before in volume. For them to bring it to market 'much quicker' than they did N3 is more than dubious. The rollout probably is not going to be substantially faster given it is substantially different (i.e., new lessons to learn). It isn't printing the '3nm' stuff at smaller '2nm' sizes. 'nm' is not what is going on here. A unidimension length isn't what is going on here with the change. It is a more complicated 3D structure , so way beyond "same stuff only smaller".

 

[Fitzman]

M4 will likely land around February 2025.

I mean it is a possibility, sure. I strongly believe Apple is wanting to get back to a one year cycle when it comes to their chips. I'm not referring to all of their devices just the more popular ones like phones & laptops mainly. Every other tech company does this, & Apple was doing this before 2020. The only reason we have the M-series at basically a 1 1/2 year gap is because of the chip shortage issue & trouble getting stuff shipped overseas.

 

[wilhoitm]

0 Nanometers (nm) = Quantum Computing 😀

 

[KOR1987]

3nm, 2nm... this is just a marketing term.

 

[Chuckeee]

What about chiplets?
AV1 hardware encoder?
TB 5?
H.266 hardware decoder/encoder?

 

[Fitzman]

What about chiplets?
AV1 hardware encoder?
TB 5?
H.266 hardware decoder/encoder?

Yeah, I'm curious about TB 5 as well. Idk if it'll happen for the M4 series or not but I'm leaning towards Apple probably won't implement that yet until maybe M5 or M6. I'm hoping that I'm wrong on this, of course. I do hope they incorporate the decoder/encoders for the M4. Thoughts?

 

[deconstruct60]

For those of us "that would be me LOL" who are not verse with these NM stuff. I found this read easy to understand.

Chip Nanometer Technology explained, and why the smaller the better

You may have spotted the term nanometer or 'nm' when reading about a chipset or going through a smartphone's specifications. But what is its significance? Does it really matter? Let's check out.

pocketnow.com

from the write up

" ... But, nowadays, the term nm isn't exactly what it used to be before. The modern-day chipset nanometer technology refers to the improvement in the manufacturing process rather than the size of the transistor itself. It has now become more of a marketing term. Nevertheless, it still reflects advancements in the industry, ... "

TSMC in more technical discussion will refer to N6 , N4 , N3 , N2 . Intel to "Intel 7" , "Intel 3" , "Intel 18A" (i.e., 1.8 Angstrom being smaller than 'nm'. ), Samsung SP4 , SF3 , etc. Long term probably more useful. It is not a direct measurement of anything on the chip. It is a more a more general indicator of some improvements in density and other multiple dimensional factors. ( more progress. )

'nm' is somewhat kept around to make longer lime comparisons to many years ago , but the fab producers are somewhat trying to move off of that term. The press and contexts where they are trying to discuss relationship to earlier products is keeping it around. There is a saying "As simple as possible and no simpler". Using 'nm' is only trying to be 'simple' ; not technically descriptive.


Additionally, at this point, to some extent it is also an indicator that the costs of making the chip are getting larger. There is also a long running assumption that future process nodes are going to make the chips much cheaper. That is somewhat evaporating also.

 

[Eric_Z]

What about chiplets?
AV1 hardware encoder?
TB 5?
H.266 hardware decoder/encoder?

Now you are talking about things that I care for. Those chiplets in particular was something I hoped we'd see for the M3.

 

[Populus]

I am not in the industry or even adjacent to it, but I love articles like that, and deep dives like Anandtech does. I watched this the other night and it blew my mind:

Thank you for sharing this video. Saving it for later.

 

[deconstruct60]

Yeah, I'm curious about TB 5 as well. Idk if it'll happen for the M4 series or not but I'm leaning towards Apple probably won't implement that yet until maybe M5 or M6.

Are you presuming that M4 comes on a yearly anniversary date? If M4 is early 2025 there is pretty good chance it is there. That is about when Intel will be in volume production of integrated TB5. No real upside in Apple getting there before Intel does ( as if Intel would sign off on that anyway. If there are not sufficient 3rd party externals to test against there is not much creditable to certify against. ).

TB5 provisions up to 3 video streams out. [ somewhat one of the primary drivers of the asymetrical 180Gbps out configuration. ] Apple is struggling with just getting to two streams out over much of the volume of systems they ship ( which as still USB4 , not TB4). If have to hit 3 out and/or DisplayPort 2.x then Apple could be in the 'hole' on meeting the Thunderbolt requirements on even more systems. Apple's display controllers run larger than 'normal' which is going to make it hard for them to crank up the number of video streams out. Pretty good chance the 'plain' M4 will be stuck on USB4 again for a fourth generation! (well perhaps USB4v2, but still not Thunderbolt ).

Modern Intel iGPUs can already routinely hit DP2.x and three output streams so they aren't going to be 'pressed' if TB has a requirement for those two things.

I'm hoping that I'm wrong on this, of course. I do hope they incorporate the decoder/encoders for the M4. Thoughts?

I doubt a AV1 encoder is coming. I suspect the AV1 encoder only came now because has 'extra space' due to the density increase of N3. N3E is as much standing for "easier" or "easing up" as it does for "enhancement". It backslides all the way back to N5 on SRAM/cache sizes. So end up with a bigger die if keep cache sizes the same (and Apple design have lots and lots and lots of on die caches. ) . So N3E likely isn't bringing capious increase in extra space if they use that. N3P wouldn't either. It would backslide less than N3E does but not really smaller.

H.266 ... if it puts more money in Apple's pocket (e.g,. contains patents that wide adoption to flow money to Apple), then maybe. Again probably depends upon available space. ( decent chance it is going to get out competed for either NPU or some incrementally more general computational increases ( apple relatively farther behind on DPv2.x adoption. AMD/intel has some.

Nvidia is slacking but probably will fix in the 5000 generation due inside of 12 months from now.

NVIDIA GeForce RTX 50 series to feature DisplayPort 2.1, using TSMC 3nm node - VideoCardz.com

NVIDIA RTX 50 with DisplayPort 2.1 support Kopite7kimi reveals further details on NVIDIA Blackwell for gamers. The RTX 40 series lacks key features like DisplayPort 2.1 and PCI Express 5.0 compatibility at launch. Both standards are already available from competing brands, such as AMD, with...

videocardz.com

).

H.266 ( VVC ) isn't getting a lot of traction. The hardware implementations are predominately decoders.

https://en.wikipedia.org/wiki/Versatile_Video_Coding#Adoption

And none of those are (very broad) mainstream usages. If there were major GPU implementors there, Apple would be 'behind', but there are not. Broadcast/streaming usages are a bit mired other cost increases; not sure many are jumping at the chance for more money to H.266 overhead also.

Apple was a very trailing edge adopter on AV1 decode.

Apple is a trailing edge adopter on DP2.x too.

When is DisplayPort 2.1 Going to be Used on Monitors? - TFTCentral

Where are all the DisplayPort 2.1 monitors? What is holding up their development and launch, and what confusion exists in the market?

tftcentral.co.uk

slow uptick on monitor makers doesn't expose slow uptick by Apple also. Multiple parts to the standards adoption chain there also.

 

[smulji]

Are you presuming that M4

This is my hunch

The new Mac mini is almost certainly coming

Apple skipped the M2 for the iMac and went straight from M1 to M3. It's possible they may do that again with M3 with respect to some Macs. I wouldn't be surprised if we see the following; Spring 2024 - M2 iPad Air / M3 iPad Pro WWDC 2024 - M4 MBA (13" / 15") Fall 2024 - M4 iMac / base MBP, M4...

forums.macrumors.com

 

[Chuckeee]

That is about when Intel will be in volume production of integrated TB5. No real upside in Apple getting there before Intel does ( as if Intel would sign off on that anyway. If there are not sufficient 3rd party externals to test against there is not much creditable to certify against. ).

TB5 provisions up to 3 video streams out. [ somewhat one of the primary drivers of the asymetrical 180Gbps out configuration. ] Apple is struggling with just getting to two streams out over much of the volume of systems they ship ( which as still USB4 , not TB4). If have to hit 3 out and/or DisplayPort 2.x then Apple could be in the 'hole' on meeting the Thunderbolt requirements on even more systems. Apple's display controllers run larger than 'normal' which is going to make it hard for them to crank up the number of video streams out. Pretty good chance the 'plain' M4 will be stuck on USB4 again for a fourth generation! (well perhaps USB4v2, but still not Thunderbolt ).

Well with announcement of usb 4 v2 to be included on razer blade 18. The push for Apple to do something beyond TB4 should be intensifying

Windows 11 will soon support USB 4 v2, the confusingly named standard that promises incredible transfer speeds

USB 4 Version 2 is on the way to PCs later this year, including the teased Razer Blade 18.

www.windowscentral.com

 

[Fitzman]

Are you presuming that M4 comes on a yearly anniversary date? If M4 is early 2025 there is pretty good chance it is there. That is about when Intel will be in volume production of integrated TB5. No real upside in Apple getting there before Intel does ( as if Intel would sign off on that anyway. If there are not sufficient 3rd party externals to test against there is not much creditable to certify against. ).

TB5 provisions up to 3 video streams out. [ somewhat one of the primary drivers of the asymetrical 180Gbps out configuration. ] Apple is struggling with just getting to two streams out over much of the volume of systems they ship ( which as still USB4 , not TB4). If have to hit 3 out and/or DisplayPort 2.x then Apple could be in the 'hole' on meeting the Thunderbolt requirements on even more systems. Apple's display controllers run larger than 'normal' which is going to make it hard for them to crank up the number of video streams out. Pretty good chance the 'plain' M4 will be stuck on USB4 again for a fourth generation! (well perhaps USB4v2, but still not Thunderbolt ).

Modern Intel iGPUs can already routinely hit DP2.x and three output streams so they aren't going to be 'pressed' if TB has a requirement for those two things.




I doubt a AV1 encoder is coming. I suspect the AV1 encoder only came now because has 'extra space' due to the density increase of N3. N3E is as much standing for "easier" or "easing up" as it does for "enhancement". It backslides all the way back to N5 on SRAM/cache sizes. So end up with a bigger die if keep cache sizes the same (and Apple design have lots and lots and lots of on die caches. ) . So N3E likely isn't bringing capious increase in extra space if they use that. N3P wouldn't either. It would backslide less than N3E does but not really smaller.

H.266 ... if it puts more money in Apple's pocket (e.g,. contains patents that wide adoption to flow money to Apple), then maybe. Again probably depends upon available space. ( decent chance it is going to get out competed for either NPU or some incrementally more general computational increases ( apple relatively farther behind on DPv2.x adoption. AMD/intel has some.

Nvidia is slacking but probably will fix in the 5000 generation due inside of 12 months from now.

NVIDIA GeForce RTX 50 series to feature DisplayPort 2.1, using TSMC 3nm node - VideoCardz.com

NVIDIA RTX 50 with DisplayPort 2.1 support Kopite7kimi reveals further details on NVIDIA Blackwell for gamers. The RTX 40 series lacks key features like DisplayPort 2.1 and PCI Express 5.0 compatibility at launch. Both standards are already available from competing brands, such as AMD, with...

videocardz.com

).

H.266 ( VVC ) isn't getting a lot of traction. The hardware implementations are predominately decoders.

https://en.wikipedia.org/wiki/Versatile_Video_Coding#Adoption

And none of those are (very broad) mainstream usages. If there were major GPU implementors there, Apple would be 'behind', but there are not. Broadcast/streaming usages are a bit mired other cost increases; not sure many are jumping at the chance for more money to H.266 overhead also.

Apple was a very trailing edge adopter on AV1 decode.

Apple is a trailing edge adopter on DP2.x too.

When is DisplayPort 2.1 Going to be Used on Monitors? - TFTCentral

Where are all the DisplayPort 2.1 monitors? What is holding up their development and launch, and what confusion exists in the market?

tftcentral.co.uk

slow uptick on monitor makers doesn't expose slow uptick by Apple also. Multiple parts to the standards adoption chain there also.

I am seeing signs of an yearly release meaning M4 would be coming at the end of this year followed by M5 near the end of 2025.

I said I doubt tb5 would come for m4 because there have been a few people who have mentioned that Apple is late in incorporating new things to their devices in a quick manner with everyone else which I believe is evident.

I don't know much on the encoders and decoders so that's why I said what I said, lol.

 

[deconstruct60]

I mean it is a possibility, sure. I strongly believe Apple is wanting to get back to a one year cycle when it comes to their chips.

"Back into" ??? When did Apple make 200-400mm^2 chips 5+ years ago. Apple didn't churn the AnnX line up for iPad Pros like that either. A10X -> A12X -> A14X . They are SKIPPING numbers there. Apple was doing one about every "full node shrink" there. ( more like a Moore's law 18 month cycle ).

The larger the chip you make the more difficult it likely is going to be to get a reasonsable return on investment(ROI) in only one year. Even the substantially much smaller A-series chips ... apple doesn't completely dump them in the trash can after only one year. There are multiple products that Apple transitions A-series chips to when 'done' with the leading edge iPhone in "hand me down fashion".

The larger the chip the vastly fewer "hand me down" targets Apple has in their line up. The plain Mx chip has multiple iPad products iPad Pro line , iPad Air to toss the chip into , but something like the Ultra does not .

That every September thing only works for the plain A-series because Apple has a wide variety of products that DO NOT update every year to divert those chips to while they play the yearly game on the bleeding edge iPhone.

Churning iPhones precisely on the same month is starting to work less also. Apple splitting the features between lead iPhone and iPhone Pro is signs of a maturing ( and slowing) market.


The Macs probably need even less churn than the iPads do.

I'm not referring to all of their devices just the more popular ones like phones & laptops mainly. Every other tech company does this,

most other tech companies have 3-10x as many products as Apple does also. Faster product release with Apple's long term support policies just get you faster de-support ( landing on Vintage/Obsolete list).

discrete GPUs don't iterate on 12 month cycles.

& Apple was doing this before 2020.

Not on iPad Pros. (which is usually about the same size die as the plain Mx is now. There is a line of sucession from A14x -> M1 )


When Apple was on Intel there is a dynamic where 40+ different system vendors who all implement systems on the same PCs are all desperately looking to differentiate themselves somehow. Throw on some some vendors trying to sell everything to everybody they are throwing out mulitple new models throughout the years. It is a medley of new systems being shipped out through most of the year. Some subset of those vendors will adopt the new chips at different rates. ( so still new places for those chips to land 2-3 years later. )

The only reason we have the M-series at basically a 1 1/2 year gap is because of the chip shortage issue & trouble getting stuff shipped overseas.

Nope. Wasn't those conditions during the A10X -> A14X cycle on bascially the same sized chip.

 

[Fitzman]

"Back into" ??? When did Apple make 200-400mm^2 chips 5+ years ago. Apple didn't churn the AnnX line up for iPad Pros like that either. A10X -> A12X -> A14X . They are SKIPPING numbers there. Apple was doing one about every "full node shrink" there. ( more like a Moore's law 18 month cycle ).

The larger the chip you make the more difficult it likely is going to be to get a reasonsable return on investment(ROI) in only one year. Even the substantially much smaller A-series chips ... apple doesn't completely dump them in the trash can after only one year. There are multiple products that Apple transitions A-series chips to when 'done' with the leading edge iPhone in "hand me down fashion".

The larger the chip the vastly fewer "hand me down" targets Apple has in their line up. The plain Mx chip has multiple iPad products iPad Pro line , iPad Air to toss the chip into , but something like the Ultra does not .

That every September thing only works for the plain A-series because Apple has a wide variety of products that DO NOT update every year to divert those chips to while they play the yearly game on the bleeding edge iPhone.

Churning iPhones precisely on the same month is starting to work less also. Apple splitting the features between lead iPhone and iPhone Pro is signs of a maturing ( and slowing) market.


The Macs probably need even less churn than the iPads do.





most other tech companies have 3-10x as many products as Apple does also. Faster product release with Apple's long term support policies just get you faster de-support ( landing on Vintage/Obsolete list).

discrete GPUs don't iterate on 12 month cycles.



Not on iPad Pros. (which is usually about the same size die as the plain Mx is now. There is a line of sucession from A14x -> M1 )


When Apple was on Intel there is a dynamic where 40+ different system vendors who all implement systems on the same PCs are all desperately looking to differentiate themselves somehow. Throw on some some vendors trying to sell everything to everybody they are throwing out mulitple new models throughout the years. It is a medley of new systems being shipped out through most of the year. Some subset of those vendors will adopt the new chips at different rates. ( so still new places for those chips to land 2-3 years later. )





Nope. Wasn't those conditions during the A10X -> A14X cycle on bascially the same sized chip

Lol, no offense but please read before you go off on me 🤣. I'm saying this respectfully (as I have been at the beginning) but Apple has indeed made laptops every year since 2010-2019. If you don't believe me, look it up.

I never mentioned anything about iPads, so there's literally no reason to bring them into the picture as they weren't on the same level. I even mentioned as you pointed out that "I am not referring to all of their devices" so not sure how that got misinterpreted.

 

[reekrish]

I suspect that, with this production timeframe, I may end up skipping the 3nm era completely. I just bought an M2 Max Mac Studio and an iPhone 15 Plus. I will probably update my iPads this year - likely new M2 Airs. I’ll be good for a while.

I was thinking the same thing.

 

[Schizoid]

I’ll just wait for 999 picometre chips, then we’ll get a 10% increase in power

 

[Chuckeee]

I’ll just wait for 999 picometre chips, then we’ll get a 10% increase in power

Heck, might as well wait until the power consumption goes negative and we can use the processors to generate electricity 😆

 

[Octavius8]

Fastest ever!
Again

 

[Tagbert]

What about chiplets?
AV1 hardware encoder?
TB 5?
H.266 hardware decoder/encoder?

In interviews, Apple tech execs have mentioned that “packaging” is an area of interest to them. That could suggest that they are considering a solution like chiplets to let them build larger SOC assemblages more flexibly and efficiently than the current method of doubling unified SOCS.

 

[giggles]

Lol, what? So I invented Coca-Cola, I guess. Just because they buy it doesn't mean they paid for its development. TSMC was gonna make a 3nm node with or without Apple, they have plenty buyers.

You drink almost 1/4 of the global Coca-Cola production?
Apple is TSMC’s biggest client at 23%.

 

[kangyula]

An 18% single core performance increase is hardly mediocre. Yes the Pro was hobbled by Apple’s reconfiguration of the chip to a true mid-range model (which affects multicore only) but an 18-20% year-over-year performance increase is nothing to sneeze at. The performance jump from Intel to M1 was a one-time thing. It’s going to be incremental gains from here on out barring a major microarchitectural shift.

Pro was downgraded not just by the loss of performance cores, but also memory bandwith. Of course that doesn't affect benchmarks that can fit into the cache, but hurts a lot of real-world applications, where the raw performance increase is actually reversed. The M2 Pro was on par with the 7800X3D, but the M3 Pro is beaten by 50%. That 20% increase only means that it waits more for memory now.

It's almost like they had a meeting, and they decided the Pro and the Max needs more separation, but both teams went and implemented their own solution, creating a Grand Canyon between the two. And now it seems as Apple forget that there exists a market outside video editors. The only flavour you can get is how fast your machine can render, no differentiation in machines based on different use cases. You want a fast CPU with quick memory access? Better pay up for bigger GPU you can't use for anything. No thanks. For that money I can build a machine with a much faster CPU (see first paragraph), a GPU that I can actually use, install an OS on it that actually can run docker smoothly without who knows how many years old bugs coming from filesystem compatibility.

That's not even the first step backwards, M1->M2 halved SSD speeds in base models. Apple managed to make me fearful instead of being excited about M4. They are missing out on a historic opportunity to increase market shares, Windows, AMD and Intel is quickly catching up, others enter with their own Arm processors. This happens when you have an old CEO who is interested in short-term gains over long-term profits.

 

[svish]

Wonder whether Apple will release a new M chip every year going forward.

 

[Fitzman]

Wonder whether Apple will release a new M chip every year going forward.

Many people including myself believe this is what Apple is working towards as there are signs of this.