Apple's First 3nm Chips for MacBook Pro Expected to Enter Production This Year

[Pilot Jones]

After "max" is "ultra". M1 Pro -> Max ->Ultra

Question is: what's next after ultra?

M2 Big Chungus

 

[deconstruct60]

if you're so inclined, read this ...

Big Changes In Architectures, Transistors, Materials

Who's doing what in next-gen chips, and when they expect to do it.

semiengineering.com

The problem several of those solutions only work on a narrow set of circuit types. The logic to implement the multiplier in a computational core is different from the logic to implement SRAM. Analog-to-digital transition logic different still.

The Expanding CXL Memory Hierarchy Is Inevitable – And Good Enough

The Hot Chips 34 conference that is normally held at Stanford University is in full swing this week, and thanks to the coronavirus pandemic is being held

www.nextplatform.com

Memory cost decreases are flattening out. That has ramifications at the on chip cache level also.

One reason why aggregated dies onto a package is where "trillion transistors on a 'chip' " is going.
If only can continue to crank up the density on only 1/4 , 1/5 , 1/8 of the die then the overall impact on overall die is going to go down on some computational problems that need the whole chip to work in a more balanced fashion.

It is more so going to open the door for more fixed function specialized compute. Certain computation problems will get way faster ( 3x - 5x ) while some 'dick , jane , spot' , extremely, general purpose compute operations do not.

Already covered in another response above. If these super exotic approaches to even smaller result in wafer costs that are 3-4x times as much as the current ones , lots of electronics designs are not going to not follow those up. There will be some that pay, but if the costs are spread over fewer and fewer players things will likely slow down over the long term.

There are not 400+mm wafers largely because nobody really wants to spend the money to covert most of the fab infrastructure over to that. If just one of TSMC , Samsung, Intel balks are going to something way more expensive then that effort has a decent chance of collapsing. Done to just one EUV fab machine maker ASML. If they don't want to do something that is way to expensive for them... it isn't happening.

 

[SoldOnApple]

The part that is more 'broken' than the density part is the prices go down part. Apple isn't primarily looking for a cheaper SoC. They are looking for an SoC that will help them hold the price points they charge. If the wafer costs go up a bit... they'll just pass those along (as much as they can ... they have large margin to work with) and keep going.

That sounds good to me. Macs will always need more power as Adobe keeps making their software more accessible, the average teenage YouTuber can count on being able to use After Effects, Illustrator, and Photoshop. Adobe will keep making more features to use that extra compute power. The base M1 Air will run all the major Adobe apps fine. Might get lag when run out of ram and scratch disk, and render time will be slow on big projects, but it will run them fine. There will never be enough computing power for what's possible and will be possible in After Effects alone.

 

[deconstruct60]

After "max" is "ultra". M1 Pro -> Max ->Ultra

Question is: what's next after ultra?

Quattro / Quattuor

or
UltraX2 ( not literally two Ultras but indicative of twice as much ).

or just plain

X ( eXtra eXtremely eXpensive. or which every one of those single one adjectives someone wants to pick. Or perhaps XL for eXtra Large. It is so bad ass it just has has a letter or two. Past words to describe it. )

or

TD ( think different. There probably is a paradigm shift here. not quite a laptop chip and not quite a legacy desktop chip either. )