After 2nm roadmap...

[sam_dean]

Source: https://www.tomshardware.com/news/i...ntil-2036-from-nanometers-to-the-angstrom-era

Check out this roadmap if you are wondering what die shrink future Apple iPhone/Mac chip will use next.

Excluding my smartphone I replace my devices after the final Security Update to the next model after it. This is typically 8-10 years apart.

So the jump in increased raw performance, increased performance per watt and decrease of power consumption would be very much apparent.

Example would be my future Macs below

- 2012 iMac 27" 22nm > 2023 iMac 27" 5nm > 2034 iMac 27" Angstrom3

- 2011 MBP 13" 32nm > 2021 MBP 16" 5nm > 2031 MBP 16" Angstrom7

This is only practical if your use case largely stays unchanged. Last change to my usage was in 2015. Since then it stayed static.

 

[MayaUser]

Since 2nm, 3nm are marketing names...after 2nm we will have 1.9nm > 1.6nm etc

 

[leman]

I really hope we’ll start moving beyond semiconductors after 2030es..

 

[jdb8167]

This might be a bit optimistic considering that with TSMC N3 & N3e static RAM (SRAM) transistors aren't shrinking. It might be too early to know if this is going to be an ongoing problem but it is worrying. A large portion of the SoC is dedicated to static RAM.

Did we just witness the death of SRAM scaling?

 

[deconstruct60]

This might be a bit optimistic considering that with TSMC N3 & N3e static RAM (SRAM) transistors aren't shrinking. It might be too early to know if this is going to be an ongoing problem but it is worrying. A large portion of the SoC is dedicated to static RAM.

Did we just witness the death of SRAM scaling?

I think those bottom two swim lanes are suppose to address aspects of that problem . 2026-28 there is an expectation of some progress , but there likely won’t be a sudden 3 year year catch up to completely close the progress gap . The gap would slowly not get much worse.

The more pressing question is how much A5 will cost per wafer in 2032. The iPhone can only get so much more expensive .The Watch even less . May get slivers of the SoC package following this path , but most of it as get farther and farther into the future has issues.

 

[deconstruct60]

Since 2nm, 3nm are marketing names...after 2nm we will have 1.9nm > 1.6nm etc

But if marketing change, why lower the percentage change. 1.9/1.6 = 1.1875 so 19% versus 9/6 = 1.5. 50% . The second is a bigger hype number to launch off of. If trying to keep hype train , then going to switch the scale.

 

[sam_dean]

Since 2nm, 3nm are marketing names...after 2nm we will have 1.9nm > 1.6nm etc

The sub-1nm nodes uses the term Angstrom. See the change of naming conventions in the photo above under the year 2024 (N2) & 2026 (A14).

Many here are amazed by the leap of performance from Intel Mac to Macs with Apple Silicon because we came from 14nm Intel chips (2014-2020) to 5nm Apple chips (2020-2023).

This leap of performance gets further magnified if you buy the replace after the final Security Update is released. To the next year model after the final Security Update. This is typically 8-10 years. Only practical if your use case does not change for that period of time.

For the purpose of parity a

- 2012 Intel Core i7 22nm 4-core CPU 77W chip

has less raw performance to a

-2023 Apple A17 Bionic 3nm >6-core CPU <6W chip

Visual indicator that would simplify this would be upgrade only when a new body Mac gets released. Like the final Intel iMac's redesign occurred in the 2012 model. Next redesign occurred in 2021. To the casual observer it appears you have the latest design without spending extra.

 

[deconstruct60]

The sub-1nm nodes uses the term Angstrom. See the change of naming conventions in the photo above under the year 2024 (N2) & 2026 (A14).

1cm and 10mm is the same thing . One uses a single digit and the other double digits.

the move to Angstrom is two fold. First, it sets up evenentually getting back to single digit again . Second , it vastly helps in not coveying more precision than is actually preset . 1.4 appears like have very accurate measurement mechanism to get to that right of the decimal point level of accuracy . There is no accuracy here . Nobody is directly measuring anything . So implying can measure the difference between 1.0 and 0.9 is an even bigger amount of hand waving than just claiming the numbers that have no direct physical instance . It is a multi dimensional metrics that is being boiled done into a single number to make it easier to demonstrate change is happening and between fab processes implementers . It isn’t accurate . It is not based on meter in the same way cm and mm are .

Many here are amazed by the leap of performance from Intel Mac to Macs with Apple Silicon because we came from 14nm Intel chips (2014-2020) to 5nm Apple chips (2020-2023).

the last upper end MBP Macs didn’t use 14nm .

 

[sam_dean]

I really hope we’ll start moving beyond semiconductors after 2030es..

It puts into perspective where we all started.

My 1st Mac was in 2000. By the 2030s it would be over 3 decades.

It would be the like looking back over 3 decades to tech or what passes as computers in the 1960s if I was still in year 2000.

If I was smart I would have sold off all devices as late as the day I received their replacements.

 

[sam_dean]

the last upper end MBP Macs didn’t use 14nm .

I had to verify that and you are correct. These were the only 10nm Macs.

- 2020 MacBook Pro 13" Core i5-1038NG7
- 2020 MacBook Pro 13" Core i7-1068NG7

But the newer 2020 iMac that were the very last Intel Macs still used 14nm Intel chips.

Too little... too late. These should have came out in back in 2016.