Apple Orders 4nm Chip Production for Next-Generation Macs

[leman]

Apple's design team being 4 times or "just" 2 times better then Intel/AMD just doesn't sound plausible

Why not? I think all the evidence points to the fact that Apple team managed to come up with a more promising microarchitecture. It didn't sprung up her night either, they've been at it for over a decade. And it's clear that they own some kind of unique recipe, since no-one else is capable of building such wide cores. The perf-per watt of Apple CPUs is unmatched both in the ARM and in the x86 space.I mean, the only other company that claims to have the comparable technology is Nuvia, which was started by former Apple's CPU designers.

And I don't think that this hypothesis is that ridiculous. It's not necessarily about some teams being better than others, it's often about the luck of having the right idea at the start. For a while, Intel had the best design around (Pentium Pro — which they still use, heavily modified, in today's CPUs) and it took AMD almost 25 years to catch up! Apple has gathered an excellent team of CPU designers, poaching a lot of talent from Intel, ARM and others, and they certainly have the resources to allow that team to go crazy. They managed to develop a much more efficient microarchitecture than anyone else is capable of, and that's where we are today.

 

[Coconut Bean]

I think you might be underestimating the microarchitecture superiority here. The 5nm will give you what, 20-30% advantage at best? And yet Apple is able to match the performance of AMD's 5nm CPUs at 1/4 power consumption. You don't get to these kind of impressive figures by process advantage alone — extremely wide backend and deep out of order execution really shine on Apple Silicon.

Amd doesn't have 5nm chips out yet..

No doubt the Apple silicon is efficient, but over time the competition will converge... Things competitors will be able to buy/copy within 1-2 generations:
* 5 nm node
* More dedicated hardware accelerators
* Memory on chip

Things they can't (cuz x86):
* Hardware/Software integration
* ARM architecture
* Most of the Apple specific improvements (large buffer, wide pipeline) to ARM Architecture

 

[tridley68]

Cool. Can't wait for I phone 13 hopefully it will get this chip

 

[Captain Trips]

Can't wait to see the upcoming Apple Silicon chips beyond M1. Good time to be a Mac owner. 👍

 

[Pimmp]

Meanwhile Intel has been struggling with 14nm chips since 2015.

Dont want to spoil the fun out of you point but just a reminder 14nm vs 14nm is not the same unfortunately.
Still Intel is about 2 nodes after competition, true story.
IBM, a boy and his atom

 

[hellosil]

If you think M1 is powerful: next year the 4nm gen will be eating it for breakfast. 🥳
Awesome stuff.

 

[wilhoitm]

We're running out of nanometres!

I wonder when we'll transition to pn (picometers) instead

There's always atoms.

"Atoms are smaller than a nanometer. One atom measures ~0.1-0.3 nm".

Help me to understand.. Aren't pico meters at the Quantum computing level?

 

[BvizioN]

Can't wait to ditch those bulky 5nm-based chips and the inefficient devices they power!

You may make a joke, but it is not too far from the truth. Technology is moving so fast that very soon your comment would seem appropriate.

 

[Captain Trips]

You may make a joke, but it is not too far from the truth. Technology is moving so fast that very soon your comment would seem appropriate.

And like I heard elsewhere (including on these forums): As good as the M1 is, it will be the slowest Apple Silicon chip released.

 

[vegetassj4]

Im waiting on the planck length chips.

 

[himanshumodi]

The silicon atom is about 0.3 nanometers; A 10 nanometers process have a transistor with about 70 atoms. I guess we are reaching the very physical limit, and apple is light years ahead intel in that space. I don’t think can miniaturise much more than probably a 1 nm. Who knows though!

Came here to find if someone was discussing physical limits.

 

[magbarn]

Good, the faster Apple moves away from 5/7nm will hopefully improve the availability of AMD 59xx CPUs, 6XXX GPU's, Xbox Xs, and PS5's! It's kind of frustrating how it seems the whole world is dependent on TSMC these days....

 

[apparatchik]

Aye but they will have their 7nm soon and Apple will be able to buy capacity there and make even more chips without relying on one supplier. And Intel 7nm != TSMC 7nm.

They key word being “soon”, so, by Intel previous promises it means “sometime before 2030, if ever”.

 

[leman]

Amd doesn't have 5nm chips out yet..

Going from 7nm (A13) to 5nm (A14) gave Apple around 20% performance improvement in SPECint, with power consumption staying essentially identical. And that is combined process improvement and microarchitecture improvement.

To achieve M1-levels of power efficiency Zen3 would need to cut down it's power consumption from 20 watts to 5 watts retaining performance. Their 5nm chips might be 25% faster or 25% more power efficient or a combination of both, but that's about it. They would need to redesign the entire CPU with power efficiency in mind which is a completely different undertaking.

 

[ruka.snow]

ok to correct your post

1. What the heck do you mean “soon”. It’s 7nm chips have been confirmed to be out in 2023 ( and most likely between now and 2023 there will be another delay) and bear in mind they were supposed to release a 7nm chip this year or last year instead of a huge delay like this (not surprising)

Soon means just that, soon. There is no hidden meaning to the word soon.

Intel's 7nm PC Chip Will Arrive in 2023 Using TSMC's Tech

The strategy allows Intel to release competitive products using manufacturing from rival foundries as it plays catch-up in the chip wars.

au.pcmag.com

and secondly do not compare Intel with TSMC either even if Intel does release 7nm in 2023, TSMC has had 7nm since 2018 and I find it much harder for Apple to buy from Intel than TSMC especially since the principal company themselves did the stupidest of all stupidity and attacked Apple with ads before offering Apple to make they’re own ARM Processors, it’s unlikely Apple will do that move

TSMC 7nm != Intel 7nm. They measure completely different things and have vastly different densities. I can sell you a shed that'll fit 7 coffee tables or another shed that will fit 7 tv units.

 

[Pimmp]

Soon means just that, soon. There is no hidden meaning to the word soon.



TSMC 7nm != Intel 7nm. They measure completely different things and have vastly different densities. I can sell you a shed that'll fit 7 coffee tables or another shed that will fit 7 tv units.

Hopefully we measure meters, decimeters, centimeters and millimeters the same for quite some time

 

[ruka.snow]

They key word being “soon”, so, by Intel previous promises it means “sometime before 2030, if ever”.

Soon has no hidden meaning. We just know that soon Apple (any many others) will have a new fab to pick from and that benefits everyone. Maybe the high end chips will go to Intel and the low too TSMC. Or maybe we'll get Intel 7nm chips in Macs at the same time as TSMC 5nm (both would be about equal).

 

[ruka.snow]

Hopefully we measure meters, decimeters, centimeters and millimeters the same for quite some time

We do and we know what a nm is. However, the measurement isn't to blame here, it is what we are measuring. Measuring MTr/mm² or µ/mm² or something else?

Intel 10nm as 100.76 MTr/mm² vs TSMC 7nm's 91.2 MTr/mm². But even then, Apple may get a less dense chip from TSMC that has better heat characteristics or win the silicon lottery with Intel and have chips that can high higher turbos, or any combination in between.

 

[macsareveryinteresting]

Intel just left the chat.

AMD Ryzen joins the chat. “Yo sup my apple fan boy dudes.” “You guys ready to have a CPU Challenge?”

 

[macsareveryinteresting]

Cool. Can't wait for I phone 13 hopefully it will get this chip

Doubt it. This Chip will be in MacBooks. And in iPhone 13, it will have the A15 Chip.

 

[Coconut Bean]

Going from 7nm (A13) to 5nm (A14) gave Apple around 20% performance improvement in SPECint, with power consumption staying essentially identical. And that is combined process improvement and microarchitecture improvement.

To achieve M1-levels of power efficiency Zen3 would need to cut down it's power consumption from 20 watts to 5 watts retaining performance. Their 5nm chips might be 25% faster or 25% more power efficient or a combination of both, but that's about it. They would need to redesign the entire CPU with power efficiency in mind which is a completely different undertaking.

TSMC quotes 30% lower power consumption at the same performance when going from 7nm to 5nm, without redesign. And there's also a density increase of 80% providing more transistor count for same die size for dedicated accelerations, architecture revamps etc.

All I am saying is that M1's dominance is a stacked effect from many smaller advantages over competitors, and that Apple hasn't design a 4x more efficient chip in general all else equal.

 

[cmaier]

Help me to understand.. Aren't pico meters at the Quantum computing level?

Not sure what your question means. I think you mean to ask about quantum effects, not quantum computing.

As it is now, there are already quantum effects in microprocessors, and have been for years. They began to get noticeable probably in the early 2000s, where they mostly manifested in so-called leakage current. This is an effect where it becomes difficult to fully turn off a field effect transistor by applying the appropriate voltage, because Schroedinger’s equation tells us that, essentially, while “the electron is most likely to be stuck in its box, there is a non-zero chance that it is sometimes outside its little box.”

This is what led Intel and others to adopt FINFETs and other structures that change the shape of the transistor to allow the electric field caused by applying a gate voltage to more easily shut off the transistor.

When people talk about the eventual demise of Moore’s law, this effect is their primary concern.

Quantum computing is a while other matter, and you generally don’t think in terms of a “size” when thinking about how to make a quantum computer.

 

[cmaier]

TSMC quotes 30% lower power consumption at the same performance when going from 7nm to 5nm, without redesign. And there's also a density increase of 80% providing more transistor count for same die size for dedicated accelerations, architecture revamps etc.

All I am saying is that M1's dominance is a stacked effect from many smaller advantages over competitors, and that Apple hasn't design a 4x more efficient chip in general all else equal.

Based on my experience, they probably get 20% over Intel (much less over AMD) from better physical design, 20% over both for RISC vs. CISC, 20% over Intel from the process advantage, 20% from micro architecture, and an amount I can’t quite put my finger on from architecture (e.g. UMA).

We did many experiments back in the day and found that our physical designs put us at least 20% ahead of Intel’s physical designs (e.g. choosing where each transistor and wire goes, how big they are, what shape they have, etc.), and the people I know at Apple are well aware of those techniques. And RISC vs. CISC advantage has been clear for years. And each time we did a process shrink we got, on average, 20% speed/watt improvement.

Weirdly, everything is always either 2% or 20% when you look at these things.

 

[iFan]

Previous macrumors articles indicated that 4nm would be used for the A16 chip in iPhone 14. I’m glad this article sort of corrects that.

What seems most likely: 5nm enhanced for upcoming iPhone. Maybe 4nm (which is really just a minor update to 5nm since you can use the same design) for upcoming macs late 2021/early 2022. And 3nm for iPhone 14 in late 2022. The iPhone 14 will not be using 4nm unless something goes horribly wrong.

 

[BeefCake 15]

Meanwhile Intel has been struggling with 14nm chips since 2015.

Yet they wanna help Apple build their chips....