Apple COO Jeff Williams Reportedly Visits Taiwan to Secure 2nm Chips

[name99]

I'm surprised Apple needs to head over to secure capacity. I mean, money talks, and Apple isn't exactly shy with their wallet when it comes to TSMC.

Even if the big picture is clear, details can always be negotiated.

"OK, you think you won't be able to begin risk production at A19 volumes until February. Fair enough. But what's the biggest bottleneck? If we wanted to run a pipecleaner M6 design in small volumes in December, could you make that happen? What would we both need to buy to get that to happen?"
etc etc.

 

[UltimaKilo]

30% further battery improvement with the new node? Holy hell, we're going to be getting close to multi-day battery life within the next 5 years.

 

[Alameda]

It’s not that easy, unless TSMC slacks off like Intel. Apple isn’t in foundry business, and there is literally no one who can match TSMC right now. Nvidia/AMD also use TSMC, in fact AMD is their second largest customer, and Nvidia isn’t too far behind. Samsung nor Intel can match up well with TSMC, unless they fumble badly in the future.

Absolutely. You can’t simply take a chip design to another chip fab house. In this case, it can only be done if TSMC builds an identical fab facility somewhere else.

 

[Alameda]

The Chief Operations Officer probably is the one that signs off in chip orders at Apple. And who's to say other engineers arent there in this case?

The Apple employees that need to be aware or, or there for, the logistics and technical peice are either already there, have been there working with TSMC for years, and or will probably be shuttling back and forth to there, unannounced over the months ahead, as they have for previous chips, and for ongoing client relations and liasing with TSMC.

Original point I was making is that the COO (or any high level Apple employee going to sign this or any other deal with suppliers or partners) going over there to officially sign the deal is largely for optics as a public /industry display of professional courtesy for its business partner.

Showing that respect still matters.

I’d imagine that COO at Apple is a lot more hands on than “Give me the document to sign.” There is much to review and negotiate on a deal that big.

 

[Alameda]

Even if the big picture is clear, details can always be negotiated.

"OK, you think you won't be able to begin risk production at A19 volumes until February. Fair enough. But what's the biggest bottleneck? If we wanted to run a pipecleaner M6 design in small volumes in December, could you make that happen? What would we both need to buy to get that to happen?"
etc etc.

For sure!!!
They don’t just say, “Gee, tell us how much it costs and we’ll just pay that much. Whatever you say…”

 

[randian]

Still waiting on a 3nm Ultra chip. Based on last week's rumors, I am guessing we get a 3nm M5 Max for the Studio and Mac Pro in 2025, but it is questionable if the Ultra still exists in their roadmap since they skipped at least one gen already. 2nm probably around M6 or so?

You think the Studio will skip both M3 and M4? I suppose updating "pro" gear every third generation is in Apple's wheelhouse given what they did to the Mac Pro for six years, but in that case why even bother having a "pro" line in the first place?

 

[dgdosen]

I have to imagine Snapdragon is already hard at work on future generations of Oryon/Elite X

 

[deconstruct60]

I have to imagine Snapdragon is already hard at work on future generations of Oryon/Elite X

Next-gen Arm Snapdragon X chips for PCs to arrive in 2025 — major Dell leak exposes Qualcomm's roadmap with 'V2' and 'V3' models

Snapdragon X Gen 2 and Gen 3 are in the cards.

www.tomshardware.com

Not only Qualcomm , but system vendors also.

the timeline at the bottom shows iterating slower than Intel . that could be influnced by how many other SoC projects Qualcomm wants them to chase with a fixed set of resources .
at some point Qualcomm will need to grow scope to doing more than one die .

 

[dgdosen]

I left out context! - I imagine they're also jetting off to Tawian to secure a foothold in that 2nm goodness. I'm sure TSMC would love a bidding war.

 

[deconstruct60]

I left out context! - I imagine they're also jetting off to Tawian to secure a foothold in that 2nm goodness. I'm sure TSMC would love a bidding war.

TSMC N2 doesn't go full production volume until 2H 2025. If Qualcomm needs to be delivering bulk Oryon v2 by that time that is a substantive problem. The build times likely are not going to be any shorter than N3 build/bake times. (i.e., need to be making in 1H to deliver in 2H. ). N2 isn't particularly viable for them (at least for these PC versions of the dies) according to that timeline. The mobile variants may have a different timeline. Qualcomm could have either targeted N3E (2H 2023) or N3P ( 2H 2024) to hit a late 2025 timeline.

pretty good chance v2 is going to be a 'clean up' that unwinds some of the server CPU core focus they probably didn't unwind so that could get the first version out. Adjust/fixes and work on that Single tread issues that pop up in single user PCs. N3P makes more sense for V2.

If look at where V3 lands in later in 2027. N2P could be in play by then ( late 2026) depending upon how much volume they need early in Q4-27 or that is mostly a slide into '28. However, getting into an "mud wrestling battle" for initial N2 wafers likely is not necessary. Qualcomm should find out more about just how many of these they are likely going to sell before getting into any bidding war on v3 production. There may be another compatible with N3E ( N3S , N3A , something that comes 'after' P. ) that works also. (Qualcomm doesn't have huge price elasticity here. Being cheaper than Intel (and AMD) is going to be a factor even at version 3. Maybe when they get to v4-6 they can ease on that. ). It would make sense to use a "mature" N2 1H 2027. The inital rollout for N3B was slow. N2 could hit the same hiccup.

If they are adding E cores with V3 or some major other 'distraction' / 'overhaul' on that iteration, then it may be worth risk mistgation to chose the N3+.. upgrade that allows more focus on updates than on design tool changes.
( a 'tock' (microarchiture) rather than a 'tick' ( fab shrink focus) ).


AMD and Intel are not going to just bail out of the laptop SoC business easily. If v3 is going to take that long to deliver there are likely going to be competition issues.

[ If Qualcomm is getting some traction on their AI 100 Inference cards and the AI/ML hype train is still looking solid. Getting some early N2 wafers for that could make sense depending where they are in the design iteration cycle. Or some kind of AI 300 UCIe chiplet by that timeframe. There is going to be lots of pressure to pull inference into the main compute die (not add in cards). That's were sooner rather than later is a more pressing issue (and has been for a couple of years). ]

 

[MacsAre1]

Bring on the Ångström era

IKR? If we're down to 2nm we'd better just start calling it 20 Angstrom.

 

[falainber]

I'm surprised Apple needs to head over to secure capacity. I mean, money talks, and Apple isn't exactly shy with their wallet when it comes to TSMC.

It's possible that other companies now may pay TSMC more for their chips than Apple. For example, NVIDIA sells their AI chips/cards for tens of thousands of dollars. They can pay way more than Apple for silicon.

 

[svish]

Apple will most likely be getting the initial batches of 2nm chips.

 

[deconstruct60]

IKR? If we're down to 2nm we'd better just start calling it 20 Angstrom.

Intel already is. For whatever reason TSMC wants to start at 16 Angstroms before switching over. ( a non clearly base 10 transition point. Perhaps 16 is more 'binar' for them. )

Joking aside in part this is because the mainstream audience is hooked to "nm" ( which has little to do with actual measuring but more of a long term logintudenal comparison metric. Back in 20xx it was 16nm so '2nm' is the same unit. avoiding unit changes for as long as possible to avoid the glazed looks when switch to a new one. )

 

[MacsAre1]

Intel already is. For whatever reason TSMC wants to start at 16 Angstroms before switching over. ( a non clearly base 10 transition point. Perhaps 16 is more 'binar' for them. )

Joking aside in part this is because the mainstream audience is hooked to "nm" ( which has little to do with actual measuring but more of a long term logintudenal comparison metric. Back in 20xx it was 16nm so '2nm' is the same unit. avoiding unit changes for as long as possible to avoid the glazed looks when switch to a new one. )

The more I look into it the more confused I get. Since the number is arbitrary and has zero to do with the size of anything.

But we changed from Khz to Mhz and Ghz, as well as KB to MB and TB, so unit changes in computing are nothing new. I'd sooner have accuracy, though.

 

[laptech]

I wonder if this is where the next anti competitive law suit will come from, other companies complaining that Apple is preventing them from using 2nm chips in their products because Apple is doing deals with TSMC to stop them from making the chips for other companies OR making sure that all the supply of 2nm chips is given to Apple and no one else.

 

[Superhai]

The more I look into it the more confused I get. Since the number is arbitrary and has zero to do with the size of anything.

But we changed from Khz to Mhz and Ghz, as well as KB to MB and TB, so unit changes in computing are nothing new. I'd sooner have accuracy, though.

It’s just marketing that keeps people talking. They could just as well used other arbitrary numbers or versions. But they use it because we are fascinated by how small it’s gotten. The same with the use of Ångström, they could just as easily use picometer. But because one sounds more sophisticated than the other, there you go.

 

[wanha]

The headline could just as easily have said:

"TSMC Secures First Order for TSMC's upcoming 2nm Chips From Apple"

Always remember the power of headlines

 

[aknabi]

Damn nice to have all the money to flex the industry… now Apple needs to use that cash to flex their software to take advantage of their hardware platform lead! Fingers crossed for an exciting WWDC!

 

[LavaLevel]

I still wonder about what's going to happen in 2025 since China has long publicly stated, that's when they WILL take over Taiwan. 😧 I was thinking about that when my 16g ram Mac Mini M2 was being shipped from there after coming from China/Hong Kong. Wish Apple would bring production, at least assembly stateside. They said they were doing that in Texas for Mac Minis for a while, but I guess that fell through.

 

[chmania]

I still wonder about what's going to happen in 2025 since China has long publicly stated, that's when they WILL take over Taiwan. 😧

Maybe, it's time to buy the M2 Mac Mini, I was planning to buy, even though there's no buying craze where I live.