Apple to occupy 90% of TSMC 3nm capacity in 2023

https://www.macrumors.com/2023/05/15/apple-tsmc-3nm-production-capacity/

by carlycue • 3 years ago

202 218 3 years ago

samwillis4• 3 years ago

Some people may criticise the "monopoly" Apply have of TSMC latest nodes, but I think it's ultimately good. TSMC need to know they have a customer for new nodes in order to invest in them, thats what Apple are in the unique position to give with their war chest of cash. This investment by Apple eventually leads its way to other customers and smaller businesses. The whole industry thrives on the wake Apple leaves behind.

As a complete other aside, I'm excited to see what comes with the M3, particularly any neural cores. I would be much more excited about all the new AI tools if they could be run locally, Apple are again in the unique position to be able to make that possible for the masses. WWDC is hopefully going to be super interesting.

What we need is a sudden and massive increase in memory on these chips to make having LLMs viable on everyday affordable devices. I do wander if that may be something Apple surprise us with.

hammock6• 3 years ago

Apple are in the unique position to give with their war chest of cash

Don’t all the big tech companies have giant cash accounts?

JKCalhoun3• 3 years ago

If I'm googling this right Apple had perhaps $48B in cash in 2022. I don't know how others compare.

reacharavindh• 3 years ago

I enjoy the competition that Apple brings to the computing market with M* chips. As an innocent bystander in the world of AI, neural networks etc, I wonder if Apple Silicon and its neural cores would become the proprietary brake like CUDA is for ML work. That is if they do something phenomenal with the 3nm facility and bring about some hardware that only they can do..

I'd rather have a healthy competetive ecosystem where people writing software need not say stuff like "needs CUDA to work or needs Apple Neural cores to work" and instead says something like "need XYZ acceleration provided by CUDA cores, Apple Silicon, and AMD blah.."

sharikous• 3 years ago

What about WebGPU? The only thing that I find unsatisfactory about it is that float64 (double) is not a standard type in WebGPU and it's not supported on Apple GPUs anyway.

audunw• 3 years ago

I think it's too early for AI/ML acceleration to standardize on an API. It takes a long time to figure out what abstractions we should standardize on before we can make a universal API.

Consider the whole history of OpenGL, Metal, Vulcan, WebGL, WebGPU...

It would have been a mistake to keep using OpenGL, and WebGL was perhaps a mistake from the beginning. It was the wrong abstractions. Metal and Vulcan are clearly better abstractions for GPU APIs, and it really doesn't matter too much that Metal is Apple-only because they're close enough that you can have a good Vulcan API on top of Metal (MoltenVK). That is, what matters is that we converge on the same abstractions, and those take a long time to get right.

Now in the end we've gotten WebGPU (which isn't only for web browsers btw, can be used from native apps too) that can provide a nice universal cross-platform API.

JKCalhoun3• 3 years ago

I suppose in some other timeline Apple funds/owns their own fab and, to the points you raise, that would be worse for everyone outside of Apple.

ftxbro• 3 years ago

It looks like the local-LLM enthusiasts will want to use Apple computers. This is funny to me for some reason, like if the most cost-effective work boot of 2025 is produced by Gucci.

foldr• 3 years ago

If you care about the combination of portability, battery life, performance, and display+audio quality, then Apple’s laptops have been price competitive for quite a while now. People who think they are overpriced usually don’t care about at least one of these things.

PragmaticPulp2• 3 years ago

Local LLMs do best on big GPUs, which aren’t an option on Macs.

What Apple has done with the M-series silicon is really impressive, but it’s not as fast as big, dedicated GPU silicon.

like if the most cost-effective work boot of 2025 is produced by Gucci.

Apple M-silicon laptops are high quality, but they’re not the same as overpriced luxury goods. The price of entry level M1/M2 Macs is extremely reasonable, IMO.

nordsieck4• 3 years ago

It is a little funny... but the combination of pretty good integrated graphics and (potentially) large amounts of on package ram mean that macs are actually a pretty inexpensive way to get access to a GPU with lots of graphics memory.

Gare• 3 years ago

Apple was cost-effective most of the time if you wanted high-quality tools.

KaoruAoiShiho• 3 years ago

I'm not aware of any local-LLM framework that works with Apple GPUs?

raxxorraxor2• 3 years ago

I think it is least likely that we will run local AI instances on Apple hardware to be honest.

But yes, the comparison would be apt. I think people want to solve a problem with dependence on third party services...

No, I expect GPUs with insane amount of memory surfacing and that isn't really a field for Apple yet.

zitterbewegung2• 3 years ago

Not many laptops have 64 gb of RAM that can run llama.cpp https://gist.github.com/zitterbewegung/4787e42617aa0be6019c3...

michaelsshaw4• 3 years ago

Nobody really benefits when the leading process node is being wasted by a walled garden company on, mostly, consumer laptops

natdempk• 3 years ago

The new nodes also typically go towards iPhones and tablets as well, and in total that's only hundreds of millions of people benefitting from improved devices, so yeah pretty much nobody.

eptcyka• 3 years ago

Laptops? Phones and tablets.

goalieca• 3 years ago

Apple is giving some pretty funding guarantees TSMC. The rest of the industry will certainly benefit.

steve_taylor• 3 years ago

It's holding back Skynet.

PragmaticPulp2• 3 years ago

Nobody really benefits when the leading process node is being wasted by a walled garden company on, mostly, consumer laptops

I would say consumers benefit if the leading process node is used on them, wouldn’t you?

I don’t understand the problem. The most efficient processes are always dedicated to mobile devices where the premium is worth it.

neogodless2• 3 years ago

There's a very common mindset that you only "benefit" if you have the latest, best "thing."

There are so many ways to benefit. I've never owned an Apple product in my life! And yet, I own an AMD CPU that was built on a TSMC process. I feel like that benefits me. Would that TSMC process be as good if they didn't have Apple as a customer? Maybe not. I don't know for sure - it's not something I obsess over.

It would be different if the products you could buy outside of Apple were actually bad. (And for some people who want fanless Apple Silicon, a lot of laptop alternatives are bad, but for my needs, both desktops and laptops with fans are fulfilling my needs and wants.)

nordsieck4• 3 years ago

What we need is a sudden and massive increase in memory on these chips to make having LLMs viable on everyday affordable devices. I do wander if that may be something Apple surprise us with.

Honestly, I just wish that they'd refresh the Mac Studio. I know that probably won't happen in any processor generation which they plan to roll out the Mac Pro, though.

apozem• 3 years ago

Exactly. TSMC can spend tens of billions of dollars on new processes because they know Apple will be there day 1 to use it all.

The iPhone and Mac have good performance and power efficiency, but people will always want their phone and laptop to do more and get longer battery life. Apple customers’ appetite for compute is practically bottomless.

reisse• 3 years ago

Not for compute, but for performance per watt. Raw compute is much better served by Intel/AMD + Nvidia workstation.

MichaelZuo5• 3 years ago

Apple customers’ appetite for compute is practically bottomless.

I don't think so, the typical iPhone user has not scaled up their compute needs much past the 2015 A9.

And if you exclude the computational photography stuff, I could imagine some users don't take advantage of it at all.

epolanski3• 3 years ago

I don't think those criticism are valid. Apple has insane order numbers, competitors do not. It's only natural they prioritize who gives them more business.

foverzar• 3 years ago

TSMC need to know they have a customer for new nodes in order to invest in them

I mean, does anyone doubt that thinner process nodes will be in demand?

samwillis4• 3 years ago

No, but companies still like to have preorders to pay for the investment, rather than having to empty their savings to raise cash. Apple with the cash they have thrown at TSMC and the almost certain high sales they know they will have, can make pre-emptive moves other companies will struggle to do.

raxxorraxor2• 3 years ago

How are we to proceed if we reach the size of a single atomic layer which is around 0.5nm?

Sure, 3nm is marketing too, but there are hard limits and reliability might suffer if manufacturing cannot keep up.

nabakin• 3 years ago

I'm excited for the M3 too. I hope this is the generation Apple adds AV1 hardware support. If they don't, I have no confidence they will ever add support

hulitu• 3 years ago

Some people may criticise the "monopoly" Apply have of TSMC latest nodes, but I think it's ultimately good.

And if Apple leaves TSMC will be left with the pants down.

thomastjeffery3• 3 years ago

Do you have any evidence whatsoever that TSMC would struggle financially without Apple?

Sounds like FUD to me.

jryle70• 3 years ago

I don't think they will be struggling, just that Apple is willing to pop up cash, or have clear commitment in advance. Businesses prefer clear and stable outlook.

Veliladon2• 3 years ago

You would be correct if investment were the bottleneck. It's not.

The bottleneck on advanced processes is how fast ASML can build its EUV fabrication machines and they're extremely complicated pieces of kit with lots of specialized parts that are also extremely complicated pieces of kit. Even with all our modern production ability we're limited in how quickly we can assemble these machines that can actually do the lithography and everybody wants them.

ASML currently has a backlog of 100 machines. Intel, Samsung, TSMC, they all want them. They sell for $200m+ a piece. Everyone has the cash for them. ASML just can't produce them fast enough.

bhouston• 3 years ago

Why can't ASML scale? They are siting on $20B of orders. That is insane. What is ASML's delivery throughput of these machines?

rjzzleep• 3 years ago

Given what has transpired about the Chips act for foreign manufacturers including TSMC I find it incredibly dumb and shortsighted of TSMC to hedge all their bets on Apple.

renewiltord2• 3 years ago

They haven't placed all their bets with Apple. Read their earnings report. They have a lot of demand. Apple is just a premier customer.

andy_ppp• 3 years ago

Is there any projection on when and if Intel will catch TSMC? Seems TSMC keep managing to deliver every single planned process improvement on time.

epolanski3• 3 years ago

None of us knows and I haven't seen any projections that Intel's gonna catch TSMC.

agd• 3 years ago

Intel claim this should be by 2025 (once their 18A process comes on line). Obviously they could miss this target, but they say they're still on track.

epolanski3• 3 years ago

They said to be on track with their 10nm node too for 8 years, how did it end?

Etheryte2• 3 years ago

Given their current pace, Intel isn't catching but falling behind instead. This has been going on for quite a long time now to be honest.

signatoremo• 3 years ago

TSMC managed to leapfrog Intel not too long ago. They still didn’t plan to adopt EUV as recently as 2017:

Not everyone is banking on EUV for 7nm, though. TSMC will extend today’s 193nm immersion and multiple patterning to 7nm, with plans to insert EUV at 5nm - [0]

Chip making is capital intensive, and a new factory that costs to the tune of $20 billions only allows a vendor to stay ahead of the competition for a few years. It’s entirely possible that Intel or Samsung will lead the race at the next technology milestone. They have work cut out for them of course. TSMC has been in the foundry business for a long time.

[0] https://semiengineering.com/why-euv-is-so-difficult/

lvl1023• 3 years ago

I feel this is a complete waste of leading tech: going into another slate of Apple products that do not need these advancements. I feel these slots should go to GPUs for AI and Apple is just hogging the capacity. It’d be one thing if Apple actually worked on AI softwares a bit and made it readily available to developers.

bombcar2• 3 years ago

If the M3 doesn’t have some AI capacities that are at least somewhat utilized by something (probably bad Siri) I’ll be highly surprised.

JKCalhoun3• 3 years ago

Everyone, please let me have "Bad Siri" for the name of my next band.

hammock6• 3 years ago

It’s the hardware version of the tech company “hire this smart person so they don’t go work somewhere else”

lvl1023• 3 years ago

It’s anticompetitive. DoJ/FTC should be all over this. They’re using leading edge for VR. Whatever they’re rolling out next month will go down as one of their biggest failures and it will be Tim Cook’s legacy.

samwillis4• 3 years ago

Stick that "GPUs for AI" in every new portable device for local efficient use and you have me won over though. I don't want my AI in the cloud, or on expensive desktop machines.

jjoonathan• 3 years ago

Apple pays a premium for the privilege of using their face to test the new node for brick walls.

Sure, sometimes it means they get the newest bestest chips a year early and the fans can party, but if you're an anti-fan just remain patient and I suspect one of these years you will be the one with reason to party.

lvl1023• 3 years ago

Which is my point exactly. They’re using market power to basically stymie competition. This is textbook anticompetitive behavior.

pjc50• 3 years ago

Apple need to make another few hundred million iPhones every year.

joakleaf• 3 years ago

Regarding this: > It’d be one thing if Apple actually worked on AI softwares a bit and made it readily available to developers.

* Apple Silicon CPUs have a Neural Engine specifically made for fast ML-inference

* Apple supports PyTorch (https://developer.apple.com/metal/pytorch/)

* Apple has its own easily accessible machine-learning framework called Core-ML (https://developer.apple.com/machine-learning/)

So it would be inaccurate to say that they are not doing anything for the ML community.

zitterbewegung2• 3 years ago

Wasn't this already announced but at 100%? https://www.extremetech.com/computing/343291-apple-has-procu... or they are just changing their order?

RcouF1uZ4gsC• 3 years ago

while Apple analyst Ming-Chi Kuo believes that new 14- and 16-inch MacBook Pro models coming in 2024 will feature M3 Pro and M3 Max chips.

I am really excited about the M3. The M1, with its unified memory it can already run decently large LLMs. It would be great if Apple was able to make the M3 even better at that. In addition, it seems like there is a lot of momentum in releasing models that the user can run on consumer hardware. I would love to have my own private LLMs running on my own hardware

justplay• 3 years ago

I was planning to purchase a Macbook, but now I think it's better to wait for M3. I am impressed with the M3 specifications. It would be ideal if I could run LLM models on my computer locally.

barbariangrunge4• 3 years ago

My impression is the m3 won’t be available until next year. Although I suspect it will be in the headset, since those things are going to need a lot of power

cainxinth• 3 years ago

Tim Cook must sweat bullets every time he reads a headline about China potentially invading Taiwan.

VWWHFSfQ• 3 years ago

Isn't USA in the process of basically relocating TSMC to Arizona for that reason? And it's not for iPhones. It's because USA military can't risk having the chip supply cut off in wartime.

dralley• 3 years ago

No. What you're missing is that the fabs moving to the US are pretty much just the ones the US military might need.

TSMC isn't moving their cutting edge production here. The fabs will be a node or two behind.

Strom2• 3 years ago

relocating TSMC to Arizona

That's a big no. What's actually happening is that TSMC is building a new factory in Arizona, including training a brand new US workforce. It remains to be seen whether it can even be done. Toyota famously failed with a similar attempt in Fremont, California.

It's a question of whether TSMC is successful because of some sort of strategy that is compatible with American work culture, or do the actual Taiwanese workers play a significant part in their success? Unanswered thus far.

dna_polymerase• 3 years ago

TSMC is expanding into the US for that reason. Their 3nm fabs will be in Arizona (in addition to Taiwan)

_zoltan_• 3 years ago

Arizona won't do N3 until 2026.

TSMC said that the leading node will always be in Taiwan and N+1 will be outside.

PaywallBuster• 3 years ago

if China invaded Taiwan, probably we'd have 100x more problems and demand for apple hardware would dwindle

prob not his worse case scenario :)

ChatGTP• 3 years ago

I'd imagine it would be WW3 and probably a civilization breaking event, would take a long time to recover from.

harveywi• 3 years ago

Sweat bullets? You may be confusing Tim Cook with Tim Gunn.

freedomben5• 3 years ago

Tim Cook has shown a pretty serious willingness to work with the CCP in the past, and their human rights track record right now is pretty bad. What makes you think either he wouldn't do so after an invasion or that China would do anything to disrupt one of the biggest prices and competitive advantages in the world (TSMC)?

adam_arthur• 3 years ago

The US Government would clearly engage in economic warfare, at the very least, with China if they were to invade Taiwan. Likely access to cutting edge chips would go with it... assuming China is occupying Taiwan.

Yes, this would lower QoL pretty significantly for both countries.

infinityplus1• 3 years ago

Who gets the remaining 10% capacity?

ChuckNorris894• 3 years ago

Sampling for future customers.

tobyhinloopen• 3 years ago

I bought it all

microtherion• 3 years ago

Bitcoin miners

0zemp2c• 3 years ago

...and when M3 laptops come to market, M1 laptops (10x the performance 95% of users will need) will drop to $499, a giant shovel of dirt spread over the Wintel coffin

cubefox2• 3 years ago

This could mean Qualcomm will release again something like the Snapdragon 888 and Snapdragon Gen 1, when they temporarily switched from TSMC to Samsung, with disappointing results.

AndrewKemendo6• 3 years ago

I’d love some super smart technologist in this forum to give a realistic and practical pathway to disrupting apple in the mobile or “computer” market.

Given the fact that Apple has such a dominant position in every part of the supply chain infrastructure I’m challenged with being able to see how any organization could possibly come up with something that pushes Apple into bankruptcy.

drexlspivey2• 3 years ago

China invades Taiwan?

AndrewKemendo6• 3 years ago

Is this legitimately the only realistic option?

jjtheblunt3• 3 years ago

Why would you want that?

AndrewKemendo6• 3 years ago

One of the hallmarks of being a hacker is a Democratic spirit. The ethos that we’re forever vigilant and deeply skeptical of structural power.

Apple has too much power, no for profit company or person can responsibly wield with that much power.

thomastjeffery3• 3 years ago

The walled garden approach devalues the already-overpriced hardware they are making. If they are taking up 90% of TSMC capacity making shit I would never buy (out of hardware I would otherwise want), I am clearly missing out on the fruits of an alternative reality.

The success of capitalism to meet the demands of a consumer are entirely dependent on competition. By vertically integrating their business, Apple has made it impossible to compete.

sosodev2• 3 years ago

Is this really something a technologist could take on? Apple could almost certainly copy and improve any hot new tech before it has time to seriously hurt their bottom line. They have a ton of resources...

My guess is that only another Apple could dethrone Apple. That's to say a company that produces high quality products with strong branding consistently over multiple decades. Google and Samsung are competitors but in my mind, as an Apple user, their products are not worthy quality wise of the luxury prices they demand.

AndrewKemendo6• 3 years ago

How would such a company start and grow without simply being absorbed or copied by Apple or another similar oligopolist?

fomine3• 3 years ago

Huawei has been great but unlikely to be next Apple due to sanction

ricokatayama• 3 years ago

And this is what we call monopsony.

neogodless2• 3 years ago

But "3nm" is not the only manufacturing service TSMC sells. They are selling from their other lines to lots of other customers.

What's the overall annual dollar amount TSMC sells, and of that, what percentage is Apple, regardless of the specific production line?

Night_Thastus• 3 years ago

The market for everything other than the cutting edge is absolutely massive. 3nm is shiny and all, but it's not what most of the world runs on.

It definitely would be nice if there were more options at the cutting edge, but it's like that for good reasons and it's not likely to change.

NelsonMinar2• 3 years ago

It's astonishing how TSMC is the only company in the world that can do this. They are just so far ahead of everyone else in manufacturing. It'd sure be nice to have more capacity.

The Economist had a good article a few months ago about the volume of chip manufacture by country and process size. The takeaway is Taiwan leads < 10nm, the US and Taiwan lead 10-22nm, and that > 22nm is an interesting mix including South Korea and China.

Chart: https://www.economist.com/img/b/400/834/90/media-assets/imag...

Article: https://www.economist.com/asia/2023/02/02/americas-hoped-for...

undersuit2• 3 years ago

You don't mention Samsung, is this the case of their 3nm not being actually 3nm?

https://wccftech.com/samsung-secures-3-nm-advanced-chip-orde...

NelsonMinar2• 3 years ago

I'm not an expert on the tech and can't say. Just going on what the Economist article talks about, South Korean manufacturing represents a tiny fraction of the smallest processes right now. The article you linked says the 3nm will start producing in 2024. That's good news, maybe there's competition coming.

apexalpha• 3 years ago

And the company that makes the lithography machine needed for theses nodes is also a monopoly... No other company in the world can do it.

It's weird to think we're dependent on single companies for such important things.

barbariangrunge4• 3 years ago

The article quotes 36gb of ram for the m3 MacBook pros. Which made me wonder: why do we use power-of-2 gb ram configurations? If apple does 12/24/36, is that reasonable, or is there a reason not to?

ac292• 3 years ago

DDR5 supports non power of 2 sized modules. You can find 48GB DDR5 modules, for example.

scrlk• 3 years ago

Apple has booked nearly 90% of chip supplier TSMC's first-generation 3-nanometer process capacity this year

No surprise, the N3B ("first gen") process has bad economics and poor yields. Everyone else is waiting for N3E, which is pretty different to N3B.

See: https://www.semianalysis.com/p/tsmcs-3nm-conundrum-does-it-e...

https://fuse.wikichip.org/news/7048/n3e-replaces-n3-comes-in...

Apple to occupy 90% of TSMC 3nm capacity in 2023

Related Stories