Intel Foundry demonstrates first Arm-based chip on 18a node
hothardware.com117 points by rbanffy 3 days ago
117 points by rbanffy 3 days ago
Assuming they’re telling the truth, they’ve successfully built one chip from that fab. That’s good, but it doesn’t mean the fab is capable of manufacturing at scale while turning a profit.
They need an external customer for the fab so they can iterate and work out the issues. It’s anyone’s guess if someone trusts intel to manufacture on their behalf instead of sticking with an established player. They’re stuck in a chicken and egg situation - can’t reach high yields without a customer, but a customer only wants to sign up if the yields and future deliveries are guaranteed.
Intels only hope might be that someone, not naming names, coerces an established company to sign up.
That's too pessimistic. In general, customers don't want to be dealing with a monopolist and foundry customers are no different. It's in everyone's interest to solve the unproven process problem, so if Intel has evidence that the process isn't bust, customers will find a product which can be used as a pipe cleaner for mutual benefit.
Specially companies like Nvidia for which the gross profit margin is so high their risk of losing TSMC is higher than risk of losing money.
Apple is similarly paranoid about single-sourcing -- off the top of my head I'm not sure whether their top-end M-class chips are currently fabbed by both TSMC and Samsung, or just TSMC>
Because if there was only a single source (for example if the other one was out-competed), they'd have to pay 30% of their revenue for the privilege of being in the FabStore.
This is already happening. The leading edge node wafers cost a fortune compared to older nodes. TSMC has limited capacity, as it takes years to bring new fabs online, and with competitors struggling they have great pricing power. Maybe why their revenue has roughly tripled over the last decade.
Samsung has already announced that their frontier node (what they call 1.4nm) is going to delayed at least two years, and issuing statements calling it into question at all. Intel has announced that they will only do what they call 14A if they can get a partner who will promise to use it in significant volume.
As of this moment, the only company that is definitely going ahead with that next generation node is TSMC. The other two companies capable of doing so are both signalling that they will only do it if they get a partner who promises to use them for significant volume, not just as negotiating leverage against TSMC.
They always are the first ones to use the most advanced node by TSMC, the designs probably are only compatible with that particular process. Have not heard of apple using samsung for SoCs.
Isn't the traditional solution to offer a really big rebate to the first customer?
Like 75% off for the first run of chips?
If Intel doesn't even want to dogfood their own node, this isn't a matter of tuning sales incentives.
https://semiwiki.com/forum/threads/nova-lake-to-use-tsmc-n2p...
Intel has a habit of giving up on things too early. So I'm not sure I would trust them with anything even if they had a better process or were less expensive or easier to work with.
Yup. Let’s see how they do with Arc. It takes multiple years and architecture revisions to catch up, and honestly they’ve been making very respectful improvements from Alchemist to Battlemage, and driver support and updates have been progressing very well.
I hope they don’t can it.
The foundries they're putting together for future manufacturing are just hoping customers will comes. Intel needs partnerships because the brand isn't the same since the core founders and builders are long gone.
I don't get it. Intel has a very huge customer for their 18A node, one that could bring billions in orders: itself.
If they themselves don't produce their chip there, why would anybody else do?
Intel certainly will use 18A for their own chips:
CEO Lip-Bu Tan: "Job number one is ramping Intel 18A at scale. Intel 18A and Intel 18A-P are critical nodes for Intel Products and will drive meaningful wafer volumes well into the next decade – starting with Panther Lake later this year."
But they don't want to be the ONLY customer. Intel wants other companies to invest, and as early in the processes as possible, so Intel doesn't have to bankroll the whole thing.
"Going forward, our investment in Intel 14A will be based on confirmed customer commitments. There are no more blank checks. Every investment must make economic sense. We will build what our customers need, when they need it"
https://newsroom.intel.com/corporate/lip-bu-tan-steps-in-the...
As far as I have read, even with themselves as the primary customer, there is still enough excess capacity to make it unprofitable to use the most advanced processes. I see it as a strict cost issue — the new fab costs $X to run. Intel can only keep it running Y% of the time with its own orders. You need someone to fill in the gap. Not to mention, at the moment the entire cost of an Intel fab is being amortized across only Intel chips. If they can spread that out to external customers, then they can start to make their CPUs more cost competitive (or better margins, or both).
Plus, if the goal is to make more chips domestically (of all kinds), Intel will need to show that they can fab chips for other customers, not just their own designs.
Here's the thing, they've completely given up and started making their (inferior to AMD) CPUs on TSMC. For example, Arrow Lake is on TSMC N3B. So it's not getting amortized over anything at all and their valuation is going to 0.
I think that's the industry's viewpoint as well. Intel's fabs' biggest customer was Intel. They're not doing well, so they're not fabbing as much especially at the leading edge. It'll death spiral.
> It’s anyone’s guess if someone trusts intel to manufacture on their behalf instead of sticking with an established player.
Intel also designs its own chips. Thus, it's hard for fabless players to buy in without worrying about their IPs being stolen. One of the strengths of TSMC is they only make chips. They don't do anything else. TSMC is highly trusted by its customers.
This is common in industry. You often do give a discount and guarantees to the first users of a system to compensate for the risk the customer is taking.
This is part of how DigitalOcean got going, Kingston gave a huge discount on a traditional HDD order if the order was switched to SSD instead because they wanted to kickstart scaled manufacturing. First time an SSD was put in and the IOPS was measured, the product direction was clear, at the time we thought it might be a CDN tho, but eventually landed on a "cloud hosting provider".
> They need an external customer for the fab so they can iterate and work out the issues.
I guess you mean Intel to iterate using its own money to get the customer's chip right, no?
that customer could've been apple. since they used to have a close relationship, till intel shit the bed.
If we assume that intel gets successful with 18A with their x86 processors, would they even have the money to finance the node after that? And the node after that which gets exponentially more expensive?
In the past x86 raked in enough money to burn a lot of it on new fab tech but non-x86 has grown immensely and floods TSMC with money. The problem for intel is that their fab tech was fitted to their processor architecture and vice versa. It made sense in the past but in the future it might not. For the processor business it may be better to use TSMC for production. For the fab it may be necessary to manufacture for many customers and take a premium for being based in a country in need. So, a split-up may be inevitable and this fabbing a competitive ARM chip surely helps in attracting more customers. Customers who may pay a premium for political and security reasons.
Apple, Nvidia and US govt can provide the required funds if they have confidence in its ability to deliver. These companies will benefit from breaking current monopoly of TSMC.
Samsung is already in a much better position for this. They have external customers and experience facilitating them. Unlike Intel's track record which doesn't inspire confidence at all.
Intel has something Samsung doesn't. It's a US company operating mostly on US soil so the US government has a vested interest to keep this strategic asset going for as long as possible.
Tech hardware is a cutthroat business, tech companies are gonna order at Intel if it has something that others don't on a business point of view: more performing, cheaper, faster delivery.
The US government can wish and encourage all they want, as long as Samsung, TSMC and any other produces better chips for less, the money will flow there.
If a government finds a sector or company to have strategic importance they will not let it die. The rest is free-market absolutism that never comes to be. I believe today more than ever the US considers Intel to be of strategic importance.
> the money will flow there
Which money? The CHIPS act [0] isn't only for the ones who produce "better chips for less".
The fact that US taxpayers will subsidize Intel does not mean that Nvidia, Google, AMD, etc are gonna other their chips there.
> Apple, Nvidia and US govt can provide the required funds if they have confidence in its ability to deliver.
Given Apple's history with Intel's ability to deliver, I'm guessing the confidence there isn't high.
Are you referring to 5G radio modems or another chip?
Probably Intel’s fumble when Apple asked them for better performance per watt for the laptop CPUs and whether they wanted the iPhone CPU business back in 2006.
A more recent motivation might be Apple's switch to in-house ARM for MacOS for similar reasons.
Well, they’re already funding so much ARM custom design, it’s not that incremental to tweak and scale for their laptops.
Probably the Intel CPUs in Macbooks before Apple made the push for the M1 - circa the Intel quad core era where their laptop chips had major heat issues... ~2012 IIRC?
I’m not defending Intel here, but those Intel MacBooks never had appropriate thermal design or headroom for the processor’s operating specs.
I think the theory is that they had an appropriate thermal design for cpus which were supposed to ship but never did.
I wouldn't count on either to save Intel as it still is (i.e with the fab business still attached to the CPU/GPU business). While it's true that having Intel fabs as a second source would be nice for them to alleviate the dependency on TSMC, they are also competing with Intel on the CPU/GPU side.
My guess is, they're gonna let Intel rot a little further while doing their best to pressure for Intel to split off their fab biz (as AMD had done back then), and then invest just in the fab.
Yep, that's exactly what they did with TSMC. Foundries don't just build massive production lines and hope someone will use them, even TSMC.
Yeah, everyone is focused on TSMC as the company with the secret sauce, but really it’s Apple. Whichever foundry Apple goes with gets the majority of leading edge transistor volume.
I understand the part where Intel is trying to get external customers interested in the output of their fab by exhibiting an implementation of an ARM processor.
In the past I understand that they did some custom implementation of Xeon cores for hyperscalers, but the meat and potatoes was the chip they designed.
Do we take this to mean that the current leadership assess the value proposition -of Intel- to be in the /making/ of the chips, akin to TSMC, and not in the /designing/ of them, as in all past seasons at Intel?
I suppose a key factor here is how far from reference this chip is. If they mean to innovate in ARM ISA territory, that's a development to ponder. But if this is a "we can also make those things" statement, I'm hearing bears in the woods.
>Do we take this to mean that the current leadership assess the value proposition -of Intel- to be in the /making/ of the chips, akin to TSMC, and not in the /designing/ of them, as in all past seasons at Intel?
No… Gelsinger laid all of this out very clearly. He wanted the design side of the house and the manufacturing side of the house to stand on their own. He didn’t want the design side relying solely on process to maintain performance leads, and he also wanted them to have the flexibility to use any fab should manufacturing fall behind.
In order for manufacturing to survive design potentially going to competitors for certain generations, they need to also support outside business.
https://www.intc.com/news-events/press-releases/detail/1451/...
The fabs need external customers not just intel to be profitable.
The custom designs for hyperscalars don’t count as external customers, they’re just part of Intels own production set.
And since nobody but AMD or VIA can make x86, it has to be ARM or other ISAs instead.
The article title is a bit clickbait since ARM is the eventuality of having external customers. The real key point is that they have made chips that aren’t their own at all.
As I understand it, Intel's strength was in manufacturing their own design in their exclusive (and most advanced) process. So the advantage was being vertically integrated. State of the art processes are too expensive these days. x86 CPUs alone cannot sustain them. Specially, when AMD builds their CPU also with state of the art processes. So by becoming a foundry, Intel may be able to have state of the art fabs and use it in their own designs of x86 CPUs, GPUs, etc.
The use of standard cells for a process somewhat opens it for outside users.
The 80386 was the first use of standard cells for x86, which also introduced "automatic place and route" via a graduate student project named "Timberwolf."
> I'm hearing bears in the woods
No, why?
The world desperately needs a TSMC competitor.
I really hope they manage to pull something out of the hat here.
Own a bunch of AMD shares so cheering for them naturally...but we don't need a monopoly in CPU space.
It would be bad for x86 in general if Intel just disappears. They supply a ton of chips for businesses still and TSMC isn't going to replace that overnight.
> Intel is effectively saying "Hey, we can make Arm chips!"
Makes sense since they were once popular in the NUC space and Apple has shown high-end ARM has a market.
Too bad they fired the ceo that made it happen
Is Intel's 18A actually 1.8 nm, or is this one of their usual marketing terms?
Process names are all marketing, at every company, not just intel. The process name has no relationship to the physical features of the transistors.
Everything for the past few generations of nodes have not been actual dimensions but more of an equivalency. TSMC nodes are no different.
Random question: where did the ARM core design come from?
Intel's first exposure was the purchase of DEC StrongARM in the 90s, although that particular product line was sold to Marvel.
Nit: Marvel makes comics. Marvell Technologies (two l's) makes chips with ARM CPUs in them, mostly for datacenter gear.
I remember one of my first PC builds had a RAID card with a Marvell controller. I can still visualize the logo on the POST screen
Intel are believed to hold an Arm architectural license [1] as far as I know, they have made Arm-based things in the past.
[1]: https://en.wikipedia.org/wiki/ARM_architecture_family#Archit...
If they didn’t have one already they would have presumably acquired one when they bought Altera - they had SoC FPGAs that have ARM cores hooked up to an FPGA fabric.
They have since spun off Altera but I imagine they’d still have a license.
I'm not sure Altera would have had an architectural license. You don't need that to hook a hard core up to your fpga fabric.
originally the MOS Technology 6502 :
https://en.wikipedia.org/wiki/ARM_architecture_family#Histor...
it's an interesting article
Probably directly from Arm? https://www.intc.com/news-events/press-releases/detail/1614/...
The actual ARM1 processor was built for the "tube" connection on a 6502-based 8-bit BBC Microcomputer in the early 1980s.
These two articles are popular for the details of that history. ARM dominates the second.
https://www.theregister.com/2012/05/02/unsung_heroes_of_tech...
https://www.theregister.com/2012/05/03/unsung_heroes_of_tech...
I'm pretty sure the grandparent's question was "What IP is on the ARM SOC being fabricated?" and not "Tell me about the history of Acorn RISC Machines".
And the answer isn't clear. The fact that it's been given an Intel code name ("Deer Creek Falls") implies that it's an internal design, so presumably it's an easily-licensed/synthesized core like a Cortex X1 or whatnot. Certainly Intel isn't expected to be designing custom ARM hardware.
It's going to be fun in two years when Intel is golden child again because TSMC has bomb damage and Taiwan is blockaded.
Or perhaps the E-Core team continues their strides and the design side becomes competitive again. AMD used to be uncompetitive after all; tides can change, and I think people are dooming too much. Intel still has a chance.
Part of Intel’s problem is their ‘P Core’ team absolutely sucked for a decade.
> TSMC has bomb damage and Taiwan is blockaded
For anyone familiar with Chinese culture, history, and mindset, and who views China through that lens rather than a Western one, the probability of this is lower than the probability of Intel’s collapsing entirely in the next two years.
“Supreme excellence is to subdue the enemy without fighting.”
“Victory without unsheathing the blade.”
“If swords are clashing, strategy has already failed.”
China is building rather innovative invasion barges.
Their plan is to invade. Or at least, that's a plan they are spending significant resources on because it's in the top five plans.
No one has doubted Intel's tech...its their manufacturing that is the problem. Anyone can make one successful chip from a wafer...making 80%+ yields is an entirely different problem to crack.
Someone moved Intel's cheese, and they didn't go after it until it was too late.
Nobody is going to be switching their ARM-based chip provider from TSMC or anyone else (with whom they've only just built up enough trust) to even thinking of changing.
Without a track record of delivery, intel is just there to be used in leverage with price negotiations with TSMC.
There’s a lot of market for ARM chips. I can totally see the likes of Mediatek giving Intel an explore if the costs are right.
why only apple and Nvidia are left buying from foundries. is the market for cpu/gpu that bad? zero innovation and other players even in niche markets?
Have you heard of AMD? You know... the company with about 25% of CPU market share (at least in PCs) these days?
https://www.tomshardware.com/pc-components/cpus/amds-desktop...
They have 17% overall according to this chart which includes Apple.
https://www.accio.com/business/best-selling-cpus
Read up on this young startup, as I think they are going places!
AMD won't buy from Intel. and I'm quoting the article.
But the article talks about Apple and Nvidia buying from Intel. You said "foundries" (plural) and clearly TSMC has other customers.
It should be RISC-V... who is in charge at Intel??
Is this related to the rumors of softbank (ARM) money injection in Intel?
From the article:
Why is Intel manufacturing an Arm SoC as a reference platform? Probably because it's trying to attract external customers, and there's a whole lot more companies building Arm SoCs than there are firms pitching x86-64 processors.
They're not trying to build the next best thing. They're trying to attract customers.
I don't think Intel plans to make a product, but to prove they can build a working chip that's not one of their own design. Being ARM has fewer developmental risks than a RISC-V design and make validation easier.
>It should be RISC-V... who is in charge at Intel??
Why should it be that? What are your arguments?
oh, you are new to HN, because you would not need to ask such question if you were reading HN in the last few years...
You can start on risc-v wikipedia page and/or on the official risc-v web site.
Very unlikely to happen but Intel could release an Arm chip with native x86 translation. Arm and AMD IP would be needed but this would be the best chip for Windows
I don't understand what the difference is between "an ARM chip with native x86 translation" and a dual-ISA x86 and ARM chip.
And I don't understand why you'd want a dual-ISA x86 and ARM rather than just an x86 chip. You wouldn't get whatever CPU front-end simplicity advantages there are from ARM, since your front-end would get significantly more complex and consume significantly more transistors than with a normal x86 chip. And I don't think there's a market of people who want ARM for compatibility reason; any Windows software which supports ARM also supports x86.
What they could do is to release an ARM chip with a slightly extended ISA to add the select features which are difficult to emulate in software, such as loads and stores with the memory ordering guarantees x86 provides but ARM doesn't. Apple does this AFAIK, and it's one part of why Rosetta 2 is so good. But any ARM CPU maker could do this.
I wonder if ARM instructions could be translated to Intel’s uOps. Then everything except that translation could be shared. And, since programs consist entirely of one type of instruction for the most part, we could imagine that the chip should be able to stick to just doing one type of translation for the duration of a program run, rather than having to figure it out for each instruction.
I’m not saying I want this, but it might be surprisingly not totally impractical.
Fujitsu and Nvidia also implement (at least) TSO.
https://threedots.ovh/blog/2021/02/cpus-with-sequential-cons...
Denver does it because it was supposed to be an x86 CPU, but they couldn't get an agreement with Intel for patent licensing, so they pivoted into being the first available aarch64 CPU since decode was happening entirely in software.
Well, it has a simple hardware decoder for what would normally be the first stage of the jit.
A chunk of what you'd want (x86 alu flag generation) seems to be an extension that is incompatible with most of the arm architectural licenses which don't allow for custom extensions to user visible space. Apple is special here for reasons that probably aren't replicable.
> I don't understand what the difference is between "an ARM chip with native x86 translation" and a dual-ISA x86 and ARM chip.
Look at Apple's Rosetta 2 for an example. M-series Apple Silicon has special undocumented modes that mirror x86 architectural quirks that don't usually exist in ARM, in order to support AOT-translated machine code. The chip doesn't support x86 instructions, but it has the amenities to support x86 code. That could be what "native x86 translation" meant?
That's what I suggested in my comment's last paragraph. I don't think that counts as "an ARM chip with native x86 translation", but really the only person who can say whether that's what dlojudice meant is dlojudice.