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.
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.
Asianometry YouTube channel did a few really good videos on chip fabbing and the unique challenges. Scale is addressed IIRC. Highly recommend.
If you have a way to churn them out please let Zeiss SMT know.
i.e. they need to project out demand x years in advance because their things are expensive and slow to build, and their components vendors have the same problem with the additional problem that their components vendors only have one customer for some of their products: them.
There is no other equivalent use for much of what they need so if they fuck it up they've got too much equipment too fast.
This is a classic hardware business problem.
Even if it was possible to double the production for free in a year then without new orders when this come online you only have 50 orders left to process ...
If you look some plane production backorders are in the decade I believe atm. Looking at the A320 orders you can see how they ramp up production (and this ramp up might production line switching from aircraft type not new lines) : https://en.m.wikipedia.org/wiki/List_of_Airbus_A320neo_famil...
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.
For processing speed, Ms are fast but I agree not anywhere near a top Nvdia chip.
For memory size, the memory on an M chip can be used as graphics memory, so a person could get an M2 today with 128GB of graphics memory for ~ $5k. Not bad considering an Nvidia chip approaching that memory size is several times that much.
It'd be pretty silly for Apple to cram in big, dedicated GPU when 99.9% of their customers don't care and don't want to pay for it, especially considering that anyone that does want big, dedicated GPU can outboard as much GPU as they want.[1] And many seem to think that the onboard GPU along with Neural Engine should be adequate for local LLM.
There is no cheaper way to have 50gb of memory allocated to your GPU compute pipeline (buy a 64gb macbook pro or studio). So yes big, only the fast aspect remains a reason to buy expensive dedicated gpu's
It's not yet. Apples combined memory architecture enables them to significantly cut the cost of bundling more memory for various GPU, Neural or other domain specific cores. I believe they will catch up or even overtake Nvidia as the leading AI platform.
Desktop AI class GPUs are hell expensive. If Apple can get something 50% as performant, but in an iPhone Pro or MacBook Air, thats going to change so very much.
Caveats: In the US, with low ram and low storage.
Without these caveats you look at about 2000€ or so (m2 macbook air with 16gb of ram and non-halved storage performance in Germany). That can still be a decent deal for what you get, but "extremely reasonable" -- not quite.
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.
[1] https://www.infoq.com/news/2023/05/mlc-llm-mobile-laptops/
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.
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.
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.."
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.
WebGL looked pretty damn good in 2011. It was based on OpenGL ES 2.0, which was the latest version of OpenGL ES, and the great thing about OpenGL ES is that it's a little cleaned up from OpenGL and reflects the capabilities of a broad class of graphics implementations.
I mean, does anyone doubt that thinner process nodes will be in demand?
Sure, 3nm is marketing too, but there are hard limits and reliability might suffer if manufacturing cannot keep up.
Don’t all the big tech companies have giant cash accounts?
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.
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.)
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.
Sounds like FUD to me.
And if Apple leaves TSMC will be left with the pants down.
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...
It's weird to think we're dependent on single companies for such important things.
https://wccftech.com/samsung-secures-3-nm-advanced-chip-orde...
I just don't think Samsung is really that far behind as right now both companies are just selling promises.
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...
prob not his worse case scenario :)
Yes, this would lower QoL pretty significantly for both countries.
TSMC isn't moving their cutting edge production here. The fabs will be a node or two behind.
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.
Every Japanese auto maker has had US plants for decades. Toyota's oldest, in Georgetown KY, has been there since 1986.
TSMC said that the leading node will always be in Taiwan and N+1 will be outside.
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.
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.
To grow without being acquired they would need to avoid funding from investors and be extremely disciplined. That's probably never going to happen in this space.
Apple has been plenty happy to work with the CCP in the past, so I don't understand how/why this would bankrupt Apple or really even cause them lots of problems.
Apple has too much power, no for profit company or person can responsibly wield with that much power.
Is it nice to have faster ARM chips? Yeah. Does it mean that Apple has a monopoly position? Not really, they just have an edge right now, I'm sure they pay eye-watering amounts to TSMC.
Intel could compete on the lower price end, but they don't, they're instead trying to convince you that they're the fastest chips on the planet, and funnel a power plants worth of electricity through the chip to get there. But they could come back to prominence with a lower power, slower, lower price chip.
I think it's an interesting question because the power you critique in Apple seems to be market share, and I'm misunderstanding how you think that is not the people voting their own opinions with their own money, thus canonically democratic.
I am sure I'm overlooking something you're thinking, so thought why not ask.
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.
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
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.
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.
What's the overall annual dollar amount TSMC sells, and of that, what percentage is Apple, regardless of the specific production line?
* 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.
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.
Using market power would be threatening TSMC to shop elsewhere if they sell the same tech to others. I’ve not heard claims they do.
I also think TSMC is the one having market power there, but given that Apple pay’s a lot for them to build the tech, that may not be as large as one would think.
So your first hit single should be about how important grass cultivation is to modern meat production: "Hay, seriously."
