Do you DARE? Europe bets once again on RISC-V for supercomputing sovereignty A 38-strong group of tech players have founded a project with the snappy name Digital Autonomy with RISC-V in Europe, aka DARE, that aims to develop processor units to power the continent’s supercomputers and other high-performance machines. If you're having deja vu, you're not alone: Europe has been talking about and toying …
I'd agree, but Europe have been very poor at owning the benefits, keeping control of technology they invent and sustaining leads they might establish. Nokia phones, Bluetooth, Cambridge Silicon, ARM, even Linux all involved European innovation that's not really lasted or isn't now European owned and controlled. You might argue Linux is open source and a lot of the intellectual muscle comes from Europe, but who actually makes money from Linux*, and why is so much of the tech world still wrapped around Microsoft? There's SAP, but that's hardly a bastion of deep skill and innovation.
Much of this is the same themes as recognised in the Draghi Report, but that excellent piece of work identified many of the problems, Europe has been far less adept at finding solutions to those problems.
* Some companies, just generally not European ones?
It is indeed good to see something like this.
I can’t help thinking though that rather than spunk millions of euros by giving it to some big tech companies to waste on management, corporate bollox, countless meetings, consultancy, legal teams, bla bla bla, they would be far better off using the money to establish a custom design house (I’m thinking Inmos-like), fill it with a couple of dozen really clever people who actually know their shit and leave them to get on with it.
The chances of success with the second strategy would be infinitely higher and you would have most of your money left-over to invest in doing the same thing with some other product
That strategy could lead to brilliant products and greater innovation, but I'm not seeing any route to commercialisation?
And that's one of the key things the Yanks have been good at. Look at the work done in Europe to create the LCD...now think who has benefited most from the LCD? Similarly the LED. Not a European invention, rather a Soviet one, but again it was the Americans who commercialised it, many decades later.
Commercialisation could come from licensing the end result to anyone who wants it. That’s what Arm have been doing for years. Or they could get the design fabbed themselves and sell the chips.
Why do you need all the big tech management/legal/etc etc to make this happen?
Come to think of it, Arm is another good example of a small team that made something really big.
For Europe, is this the things that dreams of made of, or more of a do or die effort? Is it time to open your heart to a new ISA, or has the funding only become available because of politicians listening to the sound of the crowd? Is this just seconds, because ARM and x86-64 have gobbled up all the other chip markets?
If it doesn't succeed, Europe can only blame the mirror man. (Yes, I Dare(d) to add that one in...)
"While RISC-V is open and permissively licensed, American lawmakers have sometimes called for the US to prevent China's access to the tech"
Because some US law makers are dumb as a bag of rocks, and don't understand its an open source design, so it too late to stop China using it now as they already have mass produced chips using the RISC V ISA.
Nobody said the yanks were very bright. China was importing hundreds of billions in technology from them now they're using that money to dominate them. And when they're done they will be the only country on the planet capable of fabricating chips from the ground up. Completely independent from any country.
Now after only a few years they're about to technologically surpass the US in every possible way.
From what I've understood about RISCV, it's the 2025 equivalent of the 74 series logic chips that were widely used in the 70's - in that it's a building block for a system. It's just that our idea of a building block has moved up a few levels of complexity from individual gates to units of computation.
The problem with RISCV (again, as I understand it) is that the architecture does not solve, or recreate the performance solutions found over decades of development of established architectures. You can plug in a RISCV core and it will interface with common components in a predictable way, and run much the same software as another RISCV core - but from what I've seen of commercial offerings, it will be significantly slower (perhaps orders of magnitude slower) than existing, cheap-as-chips(!) architectures.
So projects like this appear to be at a severe disadvantage... spending very significant sums to try and compete with freely available commercial devices, and hoping in n-years' time that you can narrow the performance gap from an already handicapped position. If you cannot then make it on a sufficiently modern process node within the region, any pretence at technological independence is pretty much pointless. That makes it more of a political gesture, and brings into question whether picking just a handful of companies to deliver the project is really doing much to seed innovation or independence in the wider sector.
So... hummm..
As an aside, the acronym reminds me that the Gorillaz track 'DARE' was claimed to be originally written as "There" by Albarn, but when they got Shaun Ryder in to sing, they couldn't get him to pronounce it as anything other than "Dare". Still, every project should have an anthem
RISC V is a modular instruction set, not a hardware spec. It is no different from all the other proprietary instruction sets. You take those compute function specs, design your hardware modules to execute the instructions the specs tell you to, and string 'em together on the same die. Nowadays some modules are big motherfuckers and hog too much die space. So you do the chiplet thing and physically stick the chips together in a single package. Old-style bitslice is not really a thing these days. Different builds from the same module set, mixed with a little quality grading, provide your family of products.
It's true that RISC-V performance tends to lag, but that is more to do with the sophistication and patent portfolio of the hardware design and fabrication technologies. As more players climb on board, they are driving that up hard and fast.
Well let the distinction between 74LS TTL and 4000 series CMOS logic pass us by gracefully. It's a long time since I poked a logic probe into either kind of circuit board.
Yes, RISC-V is just an instruction set architecture (ISA). But you have forgotten how open source works. There will be dozens of incompatible RISC-V extensions there. Take a look of the fragmented Linux distributions.
Do you think China, India and the rest of those seeking dominance will share their edges edges to each other as in utopia?
Early articles from the likes of the Next Platform had EPI funded SiPearl producing the HPC oriented Arm Neoverse V core Rhea chip by 2022, to be followed by Chronos. It's been put back to release this year... maybe. All very well having lots of projects but somebody somewhere needs to tape out and ship something. Happy to be enlightened as I don't follow this area terribly closely.
It may well be spreading bets, but the reality is that the ultimate "bet" is that none of this matters one iota unless one has access to TSMC's fabs in Taiwan, or a fab to match. I know that some of the critical machinery comes from the Netherlands, but there's not a competitive fab anywhere in Europe.
The US is endeavouring to persuade TSMC to set up shop in the US, but the likelihood is that that will fail. TSMC seems unwaveringly Taiwanese, and the Taiwan government has a pretty public policy of leveraging their country's mastery of silicon to keep the US from losing interest in Taiwanese defence, and for detering Chinese invasion. One Chinese solider's boot on a Taiwanese beach, and the fabs get blown sky high... My bet is that TSMC's US adventures will always be just that crucial little bit behind the leading edge.
Quite the platonic downer that EU "leadership" in supercomputing is so often "trailership" instead, as showcased here (imho). A lot of it seems to be reacting to what the rest of the world is doing (the current leader(s)) and reinventing the wheel of pommel horses that have already left the dungeon through gaping drawbridges. You can't be a leader by redoing stuff AMD, Intel, ARM, IBM, and so many others already do (chiplets, accelerators, ...) -- especially when they're industry, and you're research -- it means your research plan is researching the past rather than uncovering the stark naked future!
For the EU to be a true leader in this highly competitive and torturous space it needs to focus its medieval crossbow sights very squarely on sex! Yes, sextillions of operations per second, whipped through with the spanking satisfaction of FP128, quartered over the canopy bed of nails that will succeed PCIe 6.0 and CXL 3.+. And yes, for HPC and AI, one could use such pothole CPU arch as RISC-V, but frankly how many of the expected relevant future workloads will be of this type (as opposed to javascript and webassembly)? So why not focus on processors that are good for those rather than spray-painting wanksy graffitis over RISC-V, like lipstick on a medieval boar, basted for BBQ?
It harkens back to Hubert's comment on data sovereignty that "European governments did not have even a smidgen of vision", and to recent stoic blindness towards re-emergent geopolitcal madness, where one would have hoped for catapults of top-flight vision and planning beforehand, rather than reaction after the crisis hit the moat, much as dung does a fan ...
So do DARE EuroHPC, DARE to sex it up to the next tillion, not the previous one, your vision needs to leave gothic nostalgia behind, not re-embrace its flat flagellated butt ... and to lead from the front, shaking competitive status-quos (imo), onwards to the Zettascale!
Yeah, the CEA-Leti-coordinated Horizon Europe and Partners‘ CHIPS JU FAMES Pilot Line flagship project probably has a role to play there, expanding inwards to CPUs/SOCs from EuroServer (Leti, Chalmers, ARM, ST, OnApp, BSC, TU Dresden, Neat, Forth, and startups: ZeroPoint Tech, KALEAO). CEA-Leti (the world's premier center for advanced semiconductor solutions and olfactive force behind Aryballe’s NeOse Advance) has the 3-D integration and silicon photonics chops for this imho -- especially key for addressing memory access and celebratory champagne bottlenecks!
Then again, if the EU engages in customary "trailership", it might just end up following the fabulist Orange's deeply butt-insightful "lead" of the cognitive rectum COVID-bleach variety on this, and scuttle its own competitive advantage and CHIPS Act, in favor of drug-induced failed state diplomacy.
(We unfortunately live in contrarian times where idiocy "is" the new intelligence, and a bentover Orange "is" the new BBC, it seems!)
I wonder if we could know more about the governance model for the funding:
- is it an EU grant kinda thing where you this elaborate bureaucratic process to access funds but are hard to turf out later?
- is it a Silicon Valley (oft-intended for imitation) VC system, where you fail fast? If so, what's the oversight to protect taxpayers from abuse? (own-pocket VCs are extremely motivated)
i.e. it could be a very good idea and that's not a whole lot of money. But what's going to keep it from porking up?
Personal opinion - The EU has created an environment in which private investment for projects needing a few hundred thousand to a few millions of euros is available, but where there's no clear route to funding 100M or more privately. So it has to be the governments which fund anything of large commercial scale - however, European governments and the EU as a whole have no money and so such investment is increasingly ...difficult.
Plus, it's always a stern chase, which means it's very, very slow.
Much better to introduce a better architecture in Raspberry Pi priced product, get sales, learn, adapt and grow.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
