TSMC Expected to Begin 3nm Chip Production in Late 2022 Ahead of First M3 Macs

[Bug-Creator]

No, not simply more cores. But rather more E-cores. A different mix similar to how we see more P-cores in the M1 Pro/Max.

That would suggest doing an M2/3 variant just for the limited number of iPPs they sell, doubt that.

I also don't see the point, both iPadOS and macOS will run background tasks and some light user tasks on the E cores and once they user wants something "big" done the number of P cores define the power of the system.

Why did Apple do 4+4 in the M1? My guess is that a 1:1 ratio just seemed obvious and with 1 year of experience they may have come to the conclusion that it ain't the best and the M2 might come with just 2 E cores and the rest of the chip being used for either more P core or GPU (or something else).

The truth still is that a MBA and iPP are far close in terms of thermal/power envelope and what a user expects in terms of performance to warrant a separate chip.
Might be running at different clock or even binned one way or the other.

Pricing is a whole other issue, my iPadMini6 has the same SoC as an iPhone13Pro at half the price and I have yet to hear anybody scream "foul".

 

[nicolas17]

Friendly reminder that nanometer numbers have been complete lies for a while: https://www.pcgamesn.com/amd/tsmc-7nm-5nm-and-3nm-are-just-numbers

 

[JMacHack]

That is pure fantasy.

If you've followed chip lithography for as long as I have you'd know new 'lab' nodes are years, maybe even a decade ahead of what is commercially possible to produce. An Intel 2nm node i.e. what would be comparable to a TSMC or Samsung 2nm is half a decade away at best. Intel can't even catch up to where TSMC was with N7/N7P with their failed 10nm node, the idea they are going to leapfrog TSMC's N7, N5, N3 nodes in this next year is impossible.

Even Intel's own highly optimistic roadmaps (completely unrealistic) are lucky to have a 2nm node competitor by late 2026 and that's if everything goes perfectly and they face no delays. Realistically it's 2027 or 2028.

Intel is in 3rd place in chip manufacturing and whilst I would never write them off, I would put money on them failing to catch TSMC this decade. Given their CEO has been in talks with TSMC about using them as a fab partner tells you how little confidence Intel has in its' own manufacturing road map.

Intel's Process Roadmap to 2025: with 4nm, 3nm, 20A and 18A?!

www.anandtech.com

Is Samsungs lithography that good? I thought they were comparing to a little worse than Intel.

now I have far more power than I'll ever need.

Don’t fret, developers are hard at work making apps slower to compensate for the extra horsepower.

 

[dysamoria]

We can also assume “TSMC expected to start on 2nm production some time after quarter 2, 2022, for Apple M4...”

??‍♂️

This info is kinda useless to anyone not playing in the Wall Street gambling den.

 

[neoelectronaut]

Hmm, wonder if the "M2" is gonna make it into a 24" iMac refresh. Figure there's no sense in buying a M1 model this far past release.

 

[dysamoria]

Don’t fret, developers are hard at work making apps slower to compensate for the extra horsepower.

So much truth, yet so few recognize where the hurt comes from.

 

[dysamoria]

Sounds weird to me, M1 is being dragged to 2 years with M1 Pro and M1 Max release recently hsving atleast 1 year of shelf life. How come within 1 year they will release M2 and then go on for M3 by end of same year. I think it will make sense to have m2 in September/October as usual followed by M2 Pro and M2 Max

Probably for the same reason we are pushed toward iPhones with larger numbers every year: the numbers make people feel more pressure to buy the same stuff again, even if they don’t need (or shouldn’t need) a new device.

I can’t wait for the day when we can’t keep adding more cores and software is actually forced to be optimized...

 

[adamw]

I guess I will skip buying any more M1 and M2 (including Pro and Max variants) and wait until M3 and M3 Pro, Max, or Quadro/Quattro Macs appear. It seems that the 3nm die-shrink will be necessary for the next major version of ultimate performance to be unleashed. M2 seems like a minor speed bump to me.

 

[JPack]

That would suggest doing an M2/3 variant just for the limited number of iPPs they sell, doubt that.

I also don't see the point, both iPadOS and macOS will run background tasks and some light user tasks on the E cores and once they user wants something "big" done the number of P cores define the power of the system.

Why did Apple do 4+4 in the M1? My guess is that a 1:1 ratio just seemed obvious and with 1 year of experience they may have come to the conclusion that it ain't the best and the M2 might come with just 2 E cores and the rest of the chip being used for either more P core or GPU (or something else).

The truth still is that a MBA and iPP are far close in terms of thermal/power envelope and what a user expects in terms of performance to warrant a separate chip.
Might be running at different clock or even binned one way or the other.

Pricing is a whole other issue, my iPadMini6 has the same SoC as an iPhone13Pro at half the price and I have yet to hear anybody scream "foul".

We’ve already seen A10X and A12X just for iPad Pro. This isn’t something coming out of left field. Apple can tune transistors for lower leakage on an M2X for iPad Pro. The usage model remains very different for iPad vs. MacBook.

MBA M1 throttles by at least 10% due to lack of a fan. iPad Pro M1 doesn’t even have a heat sink. There is also a stark battery capacity difference between MacBook and iPad Pro.

I think it’s naïve to assume Apple would continue to shoehorn M-series processors into iPad Pro, especially given how quickly the segment is growing.

 

[jz0309]

No it doesn't.

One report is about a 2022 IPP which would suggest an M2, while the other tells us that 3nm will start late next year meaning product would ship in Q1 2023 or later (which would explain the "M3" part).

-> everybody is stabbing in the fog and making news out of know unknowns and unknown knowns....

yep, no-one other than TSMC know what is happening when with their 3nm, and while they keep their customers in the loop, their premier customer, won't talk about it either ...

 

[Bug-Creator]

We’ve already seen A10X and A12X just for iPad Pro.

Yep and next they put a A14x into it (everything else is just marketing).

A10x and A12x were only iPP exclusive because there was nothing else to put them in.

 

[JPack]

yep, no-one other than TSMC know what is happening when with their 3nm, and while they keep their customers in the loop, their premier customer, won't talk about it either ...

LOL, they keep their investors well informed about upcoming nodes and milestones. It’s their core business as a pure play foundry.

 

[jz0309]

LOL, they keep their investors well informed about upcoming nodes and milestones. It’s their core business as a pure play foundry.

their investor is their premier customer, Apple, or do you think Apple gets a free ride for process development?
An no, board meetings will obviously happen with some detail (and you will never hear anything about them), but no publicly traded company "informs" their investors about milestones that are not available to the general public at the same time

 

[827538]

Is Samsungs lithography that good? I thought they were comparing to a little worse than Intel.


Don’t fret, developers are hard at work making apps slower to compensate for the extra horsepower.

Samung's N3 GAA node is looking promising. I'll wait to see if they actually deliver. Right now Samsung is producing 5nm EUV FINFET chips whilst Intel is still spinning its' wheels with their disastrous 10nm process. Chip density wise it's still inferior to TSMC's N5 but better than Intel's 10nm.

 

[Realityck]

What will be smaller than 1nm.

Only that companies may begin referring to angstroms instead of nanometers, or we may simply start using decimal points.

50 A = 5 nm
10 A = 1 nm
1 A = .1 nm
.1 A = .01 nm
.01 A= .001 nm

Take a few more years to best that.

 

[You’re not me]

TSMC was sampling 2nm wafers in mid-October using a completely different process they also stated that the 4nm process would be for low cost implementations.

 

[Glideslope]

Samung's N3 GAA node is looking promising. I'll wait to see if they actually deliver. Right now Samsung is producing 5nm EUV FINFET chips whilst Intel is still spinning its' wheels with their disastrous 10nm process. Chip density wise it's still inferior to TSMC's N5 but better than Intel's 10nm.

N5 or the N5E Enhanced?

 

[sudo-sandwich]

I haven't even taken delivery of my M1 max and they're already working on its replacement. sigh.....

I'm a step further back with my recently purchased M1 mini, but it's ok. Not going to hold my breath waiting for these chips to plateau and for Apple to put the latest and greatest into the form factor I want, and buying release-date gear might be hard with supply constraints.

The base M1 is fast as heck anyway, and Wirth's Law won't catch up anytime soon because of all those Windows users on slow x86 processors.

 

[JMacHack]

Samung's N3 GAA node is looking promising. I'll wait to see if they actually deliver. Right now Samsung is producing 5nm EUV FINFET chips whilst Intel is still spinning its' wheels with their disastrous 10nm process. Chip density wise it's still inferior to TSMC's N5 but better than Intel's 10nm.

I was wondering since the RTX 3k series are notorious power hogs and it’s been said being manufactured by Samsung instead of TSMC played a part in that.

It’s not a 1:1 comparison since no x86 cpus are made on Samsungs processes. But I’ll take your word for it because I’m no expert.

 

[rgeneral]

Only that companies may begin referring to angstroms instead of nanometers, or we may simply start using decimal points.

50 A = 5 nm
10 A = 1 nm
1 A = .1 nm
.1 A = .01 nm
.01 A= .001 nm

Take a few more years to best that.

Thank you. What would be 1 atom in size. I wonder.

 

[sudo-sandwich]

Don’t fret, developers are hard at work making apps slower to compensate for the extra horsepower.

Yep, Wirth's Law is real. Luckily, cross-platform software still has to be efficient enough for all those Windows machines holding onto Intel/AMD.

 

[Epikuros]

Yep, Wirth's Law is real. Luckily, cross-platform software still has to be efficient enough for all those Windows machines holding onto Intel/AMD.

... But then they can neglect optimising the Mac version of the app, expecting AS to compensate.

 

[Seoras]

At which point does Apple hit the brick wall of die size and max cores?
I'd guess that speed becomes almost instant for basic apps and that the selling point shifts entirely to battery life.
The Mac Pro is probably too low in volume sales for it to get as much attention as the portable devices do.

 

[Seoras]

Dear CEO of INTEL. ...<snip>.. you had more than 10 years to produce a reasonable fast CPU; and all you have are lame excuses for a product which still costs top $$$. Go to the corner, think about how your failed miserably, and shut the f*ck up.

Not just reasonably fast CPU, they dropped the ball on their cellular development which caused Apple to drop its law suit against Qualcomm, pay through the nose for licensing and chips from them, buy the cellular chip development BU from Intel and do what Intel failed to do.
If they are that incompetent why would you want to buy their CPUs if you have a choice now?

 

[Flight Plan]

Wow... Kinda embarrased that I own an M1 based Mac. Nearly obsolete!

That's nothing! I won't even GET my M1 Mac until March!

How about we talk about M2 first which is not released yet.

We can't go to M2 until we have M1X first. That's what all the smart people here were saying 6 months ago.

...Don’t fret, developers are hard at work making apps slower to compensate for the extra horsepower.

Well, it's not like developers are doing it on purpose. But nobody codes in assembler anymore. We all use these huge interpreted languages that have to have massive libraries. It should be no surprise that your code is fatter than Santa Clause!

Only that companies may begin referring to angstroms instead of nanometers, or we may simply start using decimal points.

50 A = 5 nm
10 A = 1 nm
1 A = .1 nm
.1 A = .01 nm
.01 A= .001 nm

Take a few more years to best that.

Angstroms, schmangstroms. What's to stop me from referring to them as Amperage? 50 A sure looks like 50 Amps to me!