Crescent island is air cooled and likely fast enough for a group of users running something like GLM 5.2 or Kimi K3.
Druid, the 4th generation, has been shelved entirely, and all consumer DGPU products, possibly _all_ DGPU products, have been killed for Celestial.
Its likely the only products shipping with Celestial will be IGPUs for the next generation and a half, until Serpent Lake comes out with RTX graphics tiles from Nvidia, and then never another Xe product ever again.
I have no explanation other than people, researchers are comfortable with it and don't want to switch.
The hardware that nvidia is selling can be used against them because it can generate the code that is lacking for hardware from other companies.
So there’s a world out there where Intel or AMD just generate the needed software with compute from rented Nvidia hardware or the end consumer does the same, akin to opensource volunteers spending their free time reverse engineering undocumented hardware for the last few decades.
That’s if the claim that other companies produce acceptable hardware that is mostly just kneecapped by poor software.
If that’s the case there’s a distinct chance that we’ll see feature parity between the different vendors soon enough.
Unless the models that run on nvidia hardware are inadequate for this task, but that sort of raises a catch-22 — if the models aren’t good enough to generate drivers and CUDA type software what are they good enough for?
My best GPU is 5070Ti on which the best model I can run is Laguna S 2.1 118B with NVFP4 and offloading most experts on CPU with an expert router layer and 128k context.
I could kill for a 1000$ AMD or Intel card with 32 or 48GB of VRAM even GDDR6 around a 1000$ range but it feels like no one even cares.
With ~4 48GB cards I could seriously locally deploy most open models models with some SSD offloading and good 4bit quantizations with ok context sizes. But at current prices.. sigh... I tried begging Intel to maybe it's cards available in my region but alas no such luck especially not at reasonable prices.
Honestly getting an AI subscription feels cheap atm and working hard at solving and home labing stuff is insane.
Even in research no one cares about anything but Nvidia, because at this point they are expensive but they seem to care, I just don't see others caring though I have worked with Intel gpu and compute teams on implementing this stuff and they are very enthusiastic but well the companies themselves aren't solving anything for me as an individual, but they are begging for me to add support for their stack in software I maintain. The Irony is unreal, why should I even add support when no one can even use it.
Only company that had decent stuff was surprisingly Apple but their stuff is now too expensive as well. .sigh.
Also Nvidia arguably needs Intel more than vice versa, Intel now has SOTA fabs on the newest EUV technology, TSMC is at max capacity and they need to fight Apple and others for allocation...
It's an amazing piece of technology but this is... wildly wrong. ASML is a multi-national company that licenses IP largely from USA and Japan, but also Taiwan and Germany. The actual EUV light source is developed and produced in California by Cymer, which ASML acquired in 2013. But ASML was only permitted to acquire the company under a strict technology sharing and export control agreement with the US government. Additionally, a huge portion of the photolithography research is directly developed (and owned) by US companies and research organizations such as IBM, Albany NanoTech, and SEMATECH.
There is a reason why ASML's next-generation research photolithography machine is currently being installed and developed in upstate New York, and not somewhere in the Netherlands. The same reason that Cymer is still in San Diego instead of being relocated to Europe. Because it's American technology.
But Europe certainly plays a huge role in it.
https://www.governor.ny.gov/news/governor-hochul-announces-n...
https://research.ibm.com/blog/euv-center-albany-nstc
https://www.eetimes.com/asml-sematech-team-on-manufacturing-...
https://www.eetimes.com/asml-to-build-400-million-us-researc...
This is a wildly ignorant comment. The R&D lab in NY State if anything is an effort to copy IMEC, the world's leading research organization when it comes to semiconductor R&D. They essentially set and define the roadmap for node advancements and then the industry follows by actually implementing their innovations at scale in their fabs. I guess they are technically not in the Netherlands, being located about 50 miles south of the Dutch border in Belgium. And they have in fact have an EXE system of their own, and have had access to ASML's high-na tools since before they were even commercially available
They do have the xLight program which doesn’t have a true European equivalent but still is a strategic partner of ASML who have experimented with FEL in the past.
I'm not sure if it's wise to minimise the contributions of everyone else in the current climate tbh.
Why is the Netherlands government even allowing this?
That is true not only for ASML, but for the semiconductor industry in general. The number of smaller, specialized companies for certain tools, chemicals and software is staggering. And they sit all over the US, Europe, Japan, Taiwan and Korea.
Strong mutual defense agreements cover four out of the five crucial countries here (NATO, Anpo, MDT US/ROK) and the last (Taiwan) has arms sales assurances.
It’s not like these are strongly opposed nations or anything. Part of the same fleet in some sense.
The really wild thing about this whole narrative is how unlikely this entire sequence of events was. Let's review:
1) Trump gives some rambling answer to an Estonian journalist in 2017 (or 2018) in which he says about 4 different things, all of which conflict with each other.
2) That journalist clips the answer to make it sound like Trump wants to leave NATO.
3) European leaders lose their mind and for the first time in 35 years actually up their military spending. Something that every US president since Bush Sr has made a major foreign policy goal that ever single administration failed at.
4) So completely by mistake, a journalist nobody has ever heard of, handed Trump the biggest foreign policy win for any US president since Bush Sr.
5) Russia invades Ukraine
6) Sweden and Finland join NATO
7) Trump wins the 2024 election and cuts off some of the Ukrainian aid (but not intelligence sharing).
8) The removal of US limitations on Ukraine allows them to strike Russian oil infrastructure.
9) Russia can no longer afford to fund their war in the way they had been.
10) Ukraine is now actually winning.
So somehow, Trump has gotten a huge foreign policy win and helped turn around the war in Ukraine completely by accident. Its like watching a cow talk.
Cymer was one of multiple vendors (Gigaphoton, XTREME) providing this sort of light source, and ASML acquired it because it couldn't achieve what they wanted (they needed 100+W and Cymer could only achieve 15W). ASML needed to control and fund the company, and add engineering talent, to achieve what they actually needed.
Oh and ASML bought Cymer because Gigaphoton (another company at basically the same stage of development as Cymer), being a Japanese company, objected to foreign control. Though if ASML knew the protectionist, insular direction the US was going to go they probably would rethink their past decisions.
So this retconning that Cymer is really the secret magic of ASML and the Netherlands is just stealing US ingenuity is hilarious stuff.
And for that matter, it is amazing how so often we hear about "American" tech that is actually overwhelmingly combining acquisitions and manpower from abroad. Like is an iPhone an "American" creation? LOL, no.
I never said anything about ASML stealing technology or ingenuity. The point is that EUV photolithography is an international effort. The machines don't exist at all without about 6 different countries contributing technology and expertise.
ASML has certainly proven itself to be a very capable integrator of many international technologies and processes.
Why isn't there an American company doing this? Why isn't an American company making the lasers, mirrors, metrology, vacuum handling tech, etc?
Simultaneously Japan created its own initiative (EUVA), and Gigaphoton didn't rely upon the US initiative at all. And again in an alternate universe -- one where Japan isn't protectionist and Cymer wasn't out of financial runway and desperate -- we'd be talking about how ASML is really Japanese tech, etc.
If the machines are so easy to build and integrate, why isn't it happening in America?
Not to mention the critical role of the US in its development. The core EUV light source was invented in San Diego!
From ASML:
"ASML’s extreme ultraviolet (EUV) light source was developed and is manufactured at its facility in San Diego, California. Acquired from the US firm Cymer in 2013, the R&D team here developed the laser-produced plasma (LPP) system that fires lasers at tin droplets 50,000 times per second to generate EUV light."
As with most of this type of technology, most of the stuff you see is basically a giant refrigerator and giant vacuum cleaner. The actual meat is much smaller and hidden from view.
I love how ASML machines are shaped to the contours of a jet fuselage for this reason!
neither of the observations gp made is dependent upon the definition of machine, nor is it dependent on the inclusion of the examples you brought up in that definition.
It's a fantastic machine but the bulk of the achievement is clearly a joint European-American effort.
Weirdly enough it never seems to work that way in discourse.
So is this giving up on their own version? That would mean their stock price is absolutely not justified at $110, and it should be more like $60-70 if not lower. Or is this "we have technical reasons to use this. We're still going to do 18A and it's still going to beat 2nm", which would mean their stock price is easily justified at $200 and it's high time to buy.
There was a great Economist article a while ago that talks about this. ASMLs moat isn't just the know-how to build the machines themselves. It's a network of suppliers all across Europe, specialized businesses doing one little part better than anyone else in the world.
This is the type of thing that Europe is still leading in, and it'll be incredibly hard to reproduce elsewhere.
You don't just need to figure out how to build an High-NA EUV machine. You need to figure out how to build or procure all of the parts that go into it. Many of these suppliers have exclusive contracts with ASML.
Here's the article, de-paywalled: https://archive.ph/jP1HX
Of course, producing the EUV beam in the first place has also not been duplicated and it's Californian technology.
Intel 18A has a different strategy all together. I mean, it's not like they're saying much about it, but from the little we know, it's very different.
Free electron lasers are not hard at "mere EUV", and the accelerators to feed aren't either, it's just that an efficient setup requires recycling the beam which means bending it back which means a large (factory hall sized) accelerator at the beam energies needed for good EUV light.
Does this mean they do high-resolution designs in the center of the lens and low-resolution designs in the periphery?
https://www.techrepublic.com/article/news-tsmc-us-investment...
Intel staying strong is still vital for competitiveness though
https://www.cnbc.com/2021/04/21/foxconn-mostly-abandons-10-b...
In which sense? Intel is neither a full monopolist (AMD, ARM) nor has progress stalled (we're still getting a regular cycle of new generation CPUs which improve on the previous).
In a the best of timelines, intel, tsmc and samsung (and glofo maybe?) Would all have competitive fab nodes.
It also allowed AMD to completely blindside them with Zen - so we're in a situation where AMD now dominates consumer desktop and arguably a lot of enterprise/DC instead of it balancing towards a more competitive centre.
During 2012-2016 AMD was firmly in the Bulldozer disaster.
In 2017, Zen1 came out, and Intel's tick/tock mostly stopped. Zen1 was better than Bulldozer, but not better than Intel. Zen2 in 2019 was the big moment, and Intel hadn't done much. I recall an interview where the AMD person (might have been Jim Keller) said they had been targetting Zen2 to be competitive with what they thought Intel would have when it came out. Since Intel was in their pipeline bubble, Zen2 was a big win.
they might have floundered in the past.
panther lakes I hear are pretty competitive with some of the apple silicone.
Nevertheless, it is worrying that the computers with Panther Lake CPUs have become much more expensive in comparison with their equivalents of last year with Arrow Lake, and this before adding the hugely increased costs for DRAM and SSDs.
It is also confusing for buyers that the 300-series Panther Lake part numbers that appear to be replacements for the previous 200-series Arrow Lake in most cases correspond to lower technical specifications, i.e. a much smaller GPU and lower clock frequencies.
The Panther Lake models that have the new big and very good GPU are extremely expensive and also hard to find, despite the fact that the GPU continues to be produced by TSMC. Perhaps TSMC has raised its price and prioritizes other customers, so Intel cannot get enough big GPUs, but even the Panther Lake models with the very small GPU (much smaller than Intel has used in many years for its mobile CPUs) have become far too expensive.
Because Panther Lake introduced a few important new features that could have been interesting for experiments in software development (e.g. FRED), I intended to buy some mini-PC with Panther Lake, but I gave up because even one that would cost double in comparison with the mini-PC with Arrow Lake H from last year that I have (i.e. around $1000 as barebone, instead of around $500) would be a downgrade in specs, instead of being an upgrade.