Kuo: M3 Chip Still Months Away, Entering Mass Production in Second Half of 2023

[NT1440]

You really think Apple is so cheap to not redesign what’s necessary to make computer run its best? Like it was mentioned they will instead optimize any new SoC to gain performance even at some cost to energy usage.

I would think that the designs of the new machines had a few generations of M series in mind already. Tooling up tens of thousands CNC machines to build millions of units of aluminum unibody designs is a staggeringly complex process.

Apple will keep the current chassis (adding a 15” air apparently) around for at least a few more years before any major redesigns. This is the new design language for the age of Apple Silicon.

 

[Mr Rib]

Can they go sub 1 nanometer? Or is that not a thing?

-edit-
Currently, "3nm", "5nm" etc are only marketing names with no relation to the actual dimensions of the designs. Our true physical limit with the transistor architecture we are using is a quantum problem called quantum tunelling, where electrons can simply pass through fabric because of their wave like nature. If the gate in the transistor is smaller than 2 nm, (which corresponds to 5 nm design) it becomes a huge problem. Current 3nm design, physically have a gate of 48 nanometers and a tightest metal pitch of 24 nanometers ((c) wikipedia), so we are not there yet. Going beyond a 1 nm gate oxide is possible and has been done - but that's scientific experiment with carbon nanotubes, nothing suitable for mass production. Considering the numbers, we still have a lot of time before hitting the physical limit, so TSMC, Samsung and the likes have time to make a major breakthrough / invent a workaround or simply make these sub-1nm gate designs manufacturing process feasible considering the economic aspect of making them.

 

[Realityck]

From what I understand, with current architecture of a MOSFET transistor, quantum tunnelling becomes a problem at 2 nm gate oxide size. If we were to go beyond that size, we need a different, much more expensive architecture of the transistor and a change of fabric it's made of - and that can't be economically feasible solution with the manufacturing technology we have, as things are right now the costs would be astronomical. So you can say we are almost at the size limit. We either need some major breakthrough / a revolution in the approach how we create transistors or expect the processors to grow in size by binning them to get more power in foreseeable future

in the past it’s been pointed out that not all the technology is ready to continue this expectation of smaller and smaller architecture. Guess it’s time to show one of the articles again so this M3 race is not a perpetual motion thought, instead it presents technology hurdles to get by slowly.

TSMC’s 3nm: Lost in Memory Lane – Analytics India Magazine

analyticsindiamag.com

While the logical circuits have been scaling well, SRAM, the memory infrastructure, is unable to keep pace

 

[Mr Rib]

in the past it’s been pointed out that not all the technology is ready to continue this expectation of smaller and smaller architecture. Guess it’s time to show one of the articles again so this M3 race is not a perpetual motion thought, instead it presents technology hurdles to get by slowly.

TSMC’s 3nm: Lost in Memory Lane – Analytics India Magazine

analyticsindiamag.com

While the logical circuits have been scaling well, SRAM, the memory infrastructure, is unable to keep pace

One thing on that subject that comes to mind - I'm baffled and disgusted by how the software market has changed and just keeps this idiotic consumptionism wheel rolling - what happened to software optimisation? They stopped caring about software performance and threw their optimisation depts out the window - they just rely on people buying newer and newer machines, while software just sucks. I'm working with Adobe Creative Suite from like 3 years ago, why? Because new one is SO much slower, with no apparent functionality gain. Truth be told, we wouldn't even need all these new machines if developers did their work.
There are excellent examples of how good sofware could actually run - for instance iD Tech 7 engine, which runs so good mostly because of Carmack's spirit still present in the company, he is the optimisation god.
On a relatively new computer you get awesome visuals with 300 fps or so. And yet, I have a choking for last 9 years adobe illustrator just because a company making billions on subscriptions refuses to optimise / rewrite their engines with higher resolutions in mind. It really pi**es me off.

 

[steve123]

it’s been pointed out that not all the technology is ready to continue this expectation of smaller and smaller architecture

People have been forecasting the end of Moore's Law for over 60 years and it has not happened yet. For instance, look at NAND memory. Scaling the physical size of a NAND memory cell became a big problem years ago. The industry innovated in two ways. First, they started quantizing the charge level which enabled multiple bits per cell (SLC versus MLC, TLC and now QLC). Second, the began integrating them in 3 dimensions. First by chip stacking and then by layering. Expect similar innovations to appear in chip making for logic chips as well. We have a long way to go still.

 

[Biro]

You're killing me macrumors 😄. On the one hand, I desperately want the 15 inch air, but on the other maybe I should wait a few months (or year?) for the air with the M3... If it really is a significant improvement, then it's a tough call 😟

Get the 15-inch Air if you need it now. If the M3s are that great, sell or trade-in the Air next year. It won't be very old and will still be worth quite a bit. And if the 15-inch is serving all of your needs well enough, you'll have the option to wait for the M4 chip down the road. If the 15-inch Air is more of a want than a need, then hold off.

 

[Realityck]

Get the 15-inch Air if you need it now. If the M3s are that great, sell or trade-in the Air next year. It won't be very old and will still be worth quite a bit. And if the 15-inch is serving all of your needs well enough, you'll have the option to wait for the M4 chip down the road. If the 15-inch Air is more of a want than a need, then hold off.

Reminder it was back in January 2021 that the 15” MBA rumor appeared, for some reason became more talked up a year later. Now combined with M3 coming soon rumors somehow is discussed even more, but is still not anything more then an entertaining rumor.

 

[0924487]

I expect the real risk with A17 is 3nm, not the cores.

So Apple will use the massive volume of A17 to de-risk the 3nm production process and then once the iPhone production starts to ramp down towards the end of 2023, TSMC will have the fab capacity to start ramping M3 on 3nm and we will see the first M3 Macs arrive in 1Q 2024.

Can you explain what volume has to do with lowering the risk?

 

[CWallace]

Can you explain what volume has to do with lowering the risk?

The more SoCs TSMC produces, the better their quality control will be since they will be able to look for failure patterns and then develop countermeasures. So by "proofing" their process creating scores of millions of A17 SoCs to the point where they have driven down the defect rate significantly, when they start M3 production they should see much higher yields from the start thanks to said defect rate being far lower.*

* - Yes, an M3 is a more complex SoC than the A17, but it will be using the same CPU, GPU and Neural Engine cores which is likely where the failure rates are highest.

 

[_Spinn_]

I predict that we will see the M3 someday.

 

[dgdosen]

New prediction for M3s this year:

https://twitter.com/x/status/1680597241999003649

Even if all that N3B production went to the A17, I'd hope to see M3 devices no later than February. It'll be a great bellwether of Apple Silicon evolution.

 

[chucker23n1]

Almost guaranteed that Vision Pro comes with an M3.

 

[JPack]

Almost guaranteed that Vision Pro comes with an M3.

Almost zero chance. Apple has no reason to mislead developers and consumers by telling them it comes with M2.

 

[chucker23n1]

Almost zero chance. Apple has no reason to mislead developers and consumers by telling them it comes with M2.

It’s not final hardware, specs weren’t yet announced beyond a vague “it has an M2”, and developers have zero reason to worry about a newer generation. They don’t code against a microarchitecture. As for consumers, why would they mind a newer chip?

 

[Adelphos33]

It’s not final hardware, specs weren’t yet announced beyond a vague “it has an M2”, and developers have zero reason to worry about a newer generation. They don’t code against a microarchitecture. As for consumers, why would they mind a newer chip?

On the other hand, why would consumers or developers want a new chip? The difference between 1 generation in the M series is pretty minimal. Apple will want to release the Vision pro as soon as they can, and adding an M3 may delay that. M3/M4 will be in future versions of the product.

 

[JPack]

It’s not final hardware, specs weren’t yet announced beyond a vague “it has an M2”, and developers have zero reason to worry about a newer generation. They don’t code against a microarchitecture. As for consumers, why would they mind a newer chip?

The specs that have been finalized - have been announced. It's hardly "vague." Apple specifically calls out M2 in the product overview page and the press release.

Introducing Apple Vision Pro: Apple’s first spatial computer

Apple today unveiled Apple Vision Pro, a revolutionary spatial computer that seamlessly blends digital content with the physical world.

www.apple.com

Developers create apps based on the performance of M2. Apple told developers Vision Pro uses M2. Every developer will make apps that reaches the limits of M2. This is especially true given how GPU focused the headset is. Putting M3 would be wasted potential until apps get designed for M3.

Apple could have simply said "Future Apple Silicon" if they wanted developers to design apps to exceed M2 performance. Apple stated M2 because Vision Pro will come with M2.

 

[chucker23n1]

On the other hand, why would consumers or developers want a new chip?

Because it’s faster and improves battery life?

The difference between 1 generation in the M series is pretty minimal.

The M3 is likely to skip the A16 and go straight to the A17. If not the M3, then the M4; they have to skip at some point. But even if it’s A16-based, it’s likely to come in a new process node, which means more energy efficiency.

M1 to M2 wasn’t major, but would you rather not have the upgrade? Probably not.

Apple will want to release the Vision pro as soon as they can, and adding an M3 may delay that.

They’ve already announced it isn’t coming until next year.

The specs that have been finalized - have been announced. It's hardly "vague."

Very little has been finalized. It’s at least half a year away from production.

Look at the iPhone specs. In contrast, we know virtually nothing about the Vision Pro version 1 specs. Why not? Because they haven’t been finalized.

This isn’t the iPhone, where they churn out one after another year after year and need to know way in advance to get production ramped up. They’re not gonna make 200M a year of these. Maybe not even 2M in the first year.

Apple specifically calls out M2 in the product overview page and the press release.

So?

Putting M3 would be wasted potential until apps get designed for M3.

Apps do not get “designed for M3”, nor for M2. They get designed for ARM and, in the GPU case, Metal.

Apple could have simply said "Future Apple Silicon"

They could have, but “M2” is shorter and gets the “this is as powerful as a Mac” point across.

Shipping an early 2024 product with a 2022 CPU featuring 2021 cores would be silly, especially at $3500.

 

[JPack]

Very little has been finalized. It’s at least half a year away from production.

Look at the iPhone specs. In contrast, we know virtually nothing about the Vision Pro version 1 specs. Why not? Because they haven’t been finalized.

This isn’t the iPhone, where they churn out one after another year after year and need to know way in advance to get production ramped up. They’re not gonna make 200M a year of these. Maybe not even 2M in the first year.

So?

Apps do not get “designed for M3”, nor for M2. They get designed for ARM and, in the GPU case, Metal.

They could have, but “M2” is shorter and gets the “this is as powerful as a Mac” point across.

Shipping an early 2024 product with a 2022 CPU featuring 2021 cores would be silly, especially at $3500.

For any features not yet finalized, Apple would not have announced them. Did you notice how Apple did not announce the weight, RAM, storage, or colors? Those haven't been finalized. Apple announced price, processor, camera count, and battery life. Those have been finalized.

It's silly you would actually try to argue this. I suspect you've never worked for any organization with a media department. When you do a press release, that information goes out to the world. You can't click "recall." The two hour battery life and M2+R1 have been burned into the memories of millions of consumers. It's difficult to change first impressions. There is zero reason to engrave M2 into people's minds if there is no reason to do so.

The phrase "Apple Silicon" gets the point across far better than "M2" or "M3."

You're too focused on specs. Apple has been selling experience and features for the past 20 years. Vision Pro is being sold as the world's first spatial computer, not whether it has a 2022 processor or other geeky details.

 

[chucker23n1]

For any features not yet finalized, Apple would not have announced them.

But they wanted to announce that performance is at least as good as on a MacBook Air. That was the only reason they brought up the SoC at all.

It's silly you would actually try to argue this. I suspect you've never worked for any organization with a media department.

I assure you "oh, look at that, we can actually put something even better in this device than we originally teased" happens in the tech industry all the time, and is absolutely no problem at all.

The phrase "Apple Silicon" gets the point across far better than "M2" or "M3."

It doesn't. A16 is also Apple Silicon. S8 is also Apple Silicon. "M-class chip" would've been more correct, but needlessly complicated.

 

[JPack]

But they wanted to announce that performance is at least as good as on a MacBook Air. That was the only reason they brought up the SoC at all.

I assure you "oh, look at that, we can actually put something even better in this device than we originally teased" happens in the tech industry all the time, and is absolutely no problem at all.

It doesn't. A16 is also Apple Silicon. S8 is also Apple Silicon. "M-class chip" would've been more correct, but needlessly complicated.

No, Apple needed to tell developers what performance to target. Telling them to design for “Metal” as you suggest would be useless. A7 supports Metal. M2 Ultra supports Metal. Devs need to know the performance envelope.

Apple doesn’t use the term ”Apple silicon” with with A-series or Watch. Some people incorrectly do that and I’m not sure why. I guess they don’t realize there were no Intel-based iPhones or Watches. Apple Silicon A16 is redundant. That’s why Apple doesn’t ever use that.

Some tech companies change specs because they need to paper launch ahead of others to get attention. That’s not Apple.

 

[chucker23n1]

No, Apple needed to tell developers what performance to target. Telling them to design for “Metal” as you suggest would be useless. A7 supports Metal. M2 Ultra supports Metal. Devs need to know the performance envelope.

For the third time, developers do not generally target individual generations like that. They don't target an iPhone 14 differently than an iPhone 13. (They may, of course, target an iPhone 14 Pro's Dynamic island.)

As a matter of fact, App Review specifically doesn't want developers doing that, and may reject your app if you try to target a certain performance level.

The only relevance to M2 is "if it runs on your M2 Mac, then performance will be as good or better". It's not "we are putting exactly that CPU in there". If it were, they'd also have to provide how much RAM they put in, which could make quite a difference (8? 16? 24?), and which Apple did not announce.

Apple doesn’t use the term ”Apple silicon” with with A-series or Watch.

They do for developers.

Some people incorrectly do that and I’m not sure why.

Because… it's Apple Silicon?

I guess they don’t realize there were no Intel-based iPhones or Watches.

True, but the iPhone Simulator ran on Intel for more than a decade, so the architecture does play a role.

 

[JPack]

For the third time, developers do not generally target individual generations like that. They don't target an iPhone 14 differently than an iPhone 13. (They may, of course, target an iPhone 14 Pro's Dynamic island.)

As a matter of fact, App Review specifically doesn't want developers doing that, and may reject your app if you try to target a certain performance level.

The only relevance to M2 is "if it runs on your M2 Mac, then performance will be as good or better". It's not "we are putting exactly that CPU in there". If it were, they'd also have to provide how much RAM they put in, which could make quite a difference (8? 16? 24?), and which Apple did not announce.

There are over a billion iPhones in service, so of course that's the case. Vision is an entirely new computing platform. Apple needs to tell developers what performance to expect and target.

They do for developers.

Because… it's Apple Silicon?

True, but the iPhone Simulator ran on Intel for more than a decade, so the architecture does play a role.

Feel free to share a link from Apple where Apple Silicon refers to A-series or S-series chip.