Trouble Brewing for RISC-V as Issue of Technology Transfer Is Questioned | Hacker News Reader

Trouble Brewing for RISC-V as Issue of Technology Transfer Is Questioned

hackaday.com

78 pointsrcarmo49 comments

This whole thing is incredibly stupid. It's an instruction set, not some magical technology (although well designed). Transfer is the whole point: it's a hardware programming interface, an agreement between interested parties to speak the same language.

These idiots in Congress should just go ahead and ban the Chinese from using the alphabet or perhaps the digits 1 and 0.

I'm all for responding to China's increasingly aggressive posture, their challenges to a international rules bases order, their totalitarian political system, and their support of the Russian murder machine, but this is both ineffectual and wrong.

IMHO it is not stupid. It's the forever challenge of communicating with non-technical people.

These topics/questions/points are so badly written that people who understand the topic can actually make use of them to create effective rules. If you ever discussed requirements for something with non-technical people, that's the way it goes...

I'm curious to understand, on your view, what would be more effective measures?

> I'm curious to understand, on your view, what would be more effective measures?

If you absolutely wanted action, investing in US based RISC-V chips would maintain US leading edge capacity in the space.

However, arguably no action is one of the most effective options. Quite a bit of RISC-V leadership is already based (or has one foot in) the US. Between Tenstorrent via Jim Keller, Ventana, and Rivos, the leading edge designs are US based. Cutting them off from the works convinces the rest of the world to invest even harder to these American companies. Protectionism here kneecaps these US companies and creates more competitors to them.

I think there is plenty of evidence that companies can be successful and at the same time compliant in not giving strategic technological advances to oppressive regimes that will weaponise such tech.

I think for leading edge technology, if these regimes can reproduce your work at all, whether they do so is more about the incentive structures in place. By restricting access, you provide the incentives for them to truly indigenously invest in the technology. By restricting access, you actually remove all control of the technology you had in the first place.

After a bunch of back and forth, the US blocked access to Intel chips used in the Sunway Tianhe-2 supercomputer. The next version, the Sunway TaihuLight, was the fastest supercomputer in the world for two years, running on indigenously designed Chinese chips.

Instruction set specs are not a strategic technological advance that can be weaponized. They just say what a chip should do. They don't help you make one that's more sophisticated than a basic unoptimized microcontroller, and "basic unoptimized microcontroller" is something you can get random college graduates to do.

Effective to what end?

"It's illegal to talk about certain things, but only if you know Chinese nationals are listening" is an insane position to take, especially in the context of the US's otherwise extremely liberal approach to free speech.

Or are we arguing that people's private work is in fact some kind of state secret?

ITAR and EAR would like a word.

"It's illegal to talk about certain things, but only if you know Chinese nationals are listening" has applied to a lot of things for a long time.

There was a case a while back when a satellite guy answered a question about power requirements for a cubesat while overseas for the launch. When he got home the FBI talked to him about violating ITAR. I think he was able to work it all out with the application of multiple tens of thousands of dollars in legal fees.

But the first rule of ITAR club is you do not talk about ITAR club related technical details when at a launch facility in southern Russia.

I think the existing measures are well targeted and effective. There's no point trying to stop people from participating in standards, research or discussions, since they're all inherently open.

But restricting access to EUV machines and other proprietary high performance implementations makes sense.

Of course both of those are independent of ISA.

Is it about the instruction set or the reference implementations?

When they say "RISC-V", they're not talking about the ISA, but all the tech that goes along with making a core with things like the Rocket or BOOM generators. I suspect they're dissing on SiFive because they created a proprietary derivative of the Rocket core generator, transferred it to a Chinese subsidiary (StarFive), and then couldn't compete with StarFive because SiFive was mostly staffed by academics and then had the temerity to not be bought by Intel.

I had always thought SiFive management got approval for the tech transfer (you're supposed to give the State Department / FISA Courts the right to deny the transfer of tech overseas.) I don't believe State or Commerce had a problem with what SiFive was exporting at the time.

My guess is some republicans in congress found out the Trump Administration allowed the export of technology a Chinese subsidiary did something cool with and that just rubs them the wrong way, so why not blame a Democrat?

So yes... there are corners of congress that are hilariously uninformed, but it's not quite as bad as you make out.

And let's not forget that some people mean "The RISC-V Foundation" when they say "RISC-V". I think my point is human language is often ambiguous and it's entirely possible that when a congressional staffer says "RISC-V", they're not using it in the same sense as most of the rest of the tech community.

> I'm all for responding to China's increasingly aggressive posture, their challenges to a international rules bases order, their totalitarian political system, and their support of the Russian murder machine, but this is both ineffectual and wrong.

You are competitive, we get it. But can you at least stop turning every conflict of interest you have into a f-ing holy war while completely denying any agency for yourself?

Have you considered that this way you'll might stop finding yourself in situations like these?

Like, maybe if someone didn't treat China aggressively in the first place, they wouldn't have to experience pushbacks from it - yeah, what else could have happened? Maybe "Russian murder machine" (lol, how well that had aged) wouldn't be running if someone didn't lie to it about peace agreements, preparing to warfare with it instead - what else could have happened? And these countries are hardly alone in an increasing and diverse list of US enemies - there is a clear pattern with only really one thing (or actor) in common.

I have no idea what you people see behind the "international rules bases order" propaganda meme that you started massively parrot recently, but try giving that speech to someone from southern hemisphere.

There is a better way to live than treat everything around you with "barbarians at the gate" attitude.

https://selectcommitteeontheccp.house.gov/sites/evo-subsites...

This is the letter and it's as stupid as you imagine.

Well geopolitically it makes sense to some degree. Probably no other country would benefit more from RISC-V catching up with ARM/AMD64 than China (and unlike those US won't have as much direct control over transfer/licensing of IP etc.)

They seem to be extremely clear about motivations:

> The intellectual property used to design semiconductor chips is currently dominated by western firms, including ARM, Intel, and AMD. The U.S. and other governments have typically required licenses for certain use and users of chip architecture, but RISC-V allows the PRC to use open-source architecture to develop advanced chips without needing a license from the U.S. government

One might justifiable strongly disagree with that, I wouldn't it's "stupid", though.

The US should make sure it has local fabs or fabs in very friendly countries that can meet demand, but the idea of preventing China from having its own cutting edge fabs seems completely absurd and impossible.

That's a strawmen argument. It's not about preventing something but about delaying it or increasing the cost of it. Economic warfare.

Can you expound on that?

> RISC-V allows the PRC to use open-source architecture to develop advanced chips without needing a license from the U.S. government

I'm glad you highlighted that sentence from the statement because it is wrong in just about every aspect.

Firstly the US government doesn't license chip designs. Secondly there are plenty of open architectures already (Loongson being just one). Thirdly the PRC doesn't need to copy an open source architecture, they could copy x86 without licensing if they wanted to. Fourthly it's open source so how would you realisticly prevent the PRC from getting it? - Even trying will mainly hurt developers in the US and make no difference for China.

So yes, stupid.

Edit: Although it can be both stupid and dangerous when you've got US representatives writing such rubbish. That's the worst kind of stupid unfortunately.

> Firstly the US government doesn't license chip designs

I'm not sure if 'license' is the right word, maybe not, however the US government does pretty tightly control what technologies you're allowed to license/sell to foreign entities.

> Secondly there are plenty of open architectures already (Loongson being just one)

If Western companies don't invest into RISC-V that would slow down any meaningful progress in the ecosystem. Of course it might also mean that they will be left behind by Chinese companies.

> they could copy x86 without licensing if they wanted to.

They could, good luck selling it to anyone outside China. Also Chinese companies which did that would probably sanctioned to such an extent that they couldn't sell any of their products in half of the world (not just chips).

> Although it can be both stupid and dangerous when you've got US representatives writing such rubbish

Again. How is it stupid? I mean it's a blatant attempt at protectionism and (I'm not sure what the right word for that is, but something related to imperialism?). I don't think it's irrational, though. Unless you don't believe that RISC-V becoming more widespread posses a threat to the hegemony US/Western companies have in the highend chip market?

It's stupid because RISC-V isn't a specific new technology, something like an EUV machine. You can control where those are shipped. It's also not really a new idea, like patenting a new method for speculative execution or somesuch. You could work with US industry to keep those a trade secret.

RISC-V is an attempt to make a pretty boring but cross-industry standard architecture that doesn't need licensing. The fact that it doesn't need licenses is the giveaway that RISC-V isn't groundbreaking exciting new work, but more like Linux, a dull, but free, reimplementation of an existing idea.

Ask yourself if the US can control Linux distribution, or if it would have been a good idea back in 1991 for the US government to support proprietary Unix and prevent US developers from working on Linux or BSD. How would that have gone do you think? Would we be in a better place now if US persons could not work on Linux and if the US was funding SunOS development?

(And I say this as a big fan of RISC-V and the first person to work on the Fedora port to RISC-V).

> RISC-V is an attempt to make a pretty boring but cross-industry standard architecture that doesn't need licensing

Exactly. Why would the US government want that when they have significant influence over the companies(y) that you currently have to license this tech from?

> Unix and prevent US developers from working on Linux or BSD. How would that have gone do you think?

Just to clarify I'm not saying I support this, it just seems to me that what are they doing makes sense from a certain perspective. Having said that I don't believe opensource HW can really work that well (compared to software at least). Linux et al. succeeded the most (or maybe only) in areas where software is a cost center and not the actual end product. If they adopt OSS software most companies don't have much to loss and usually have something to gain by contributing to it.

I just don't see that working in the mid/high-end chip market (we seem to agree somewhat on that, though). If you're in the business of designing (not manufacturing) chips what possible incentives can you have to open your designs? Even licensing doesn't seem like a good option (look at the peanuts ARM is making). You'd only turn your product into a commodity and nobody wants that because it means lower margins.

Instead you'd try to corner the market and then add proprietary extensions & other stuff to lock your users in (and having a third party like ARM which at least levels the field somewhat seems like a preferable options). Of course you might end up with entire business model disappearing, just like proprietary UNIX etc., due to similar reasons and being replaced by fab/cloud/consumer tect/etc. companies funding OS designs instead.

I totally understand what the representatives are trying to do, but it's not going to work because they fundamentally don't understand the way that things work. That's a recipe for stupid and dangerous rules that will backfire on the US.

> I don't believe opensource HW can really work that well

It'll work because hardware vendors are always trying to reduce costs, and licensing costs are part of that. A big part for embedded designs, less so for high end stuff, and guess where RISC-V is making inroads right now.

> Instead you'd try to corner the market and then add proprietary extensions & other stuff to lock your users in

Definitely a danger that this could happen. At Red Hat we've been pretty clear that we wouldn't support vendors that do this.

To be clear the letter doesn’t say that the US should support proprietary ISAs, in fact it doesn’t even say US persons cannot work with someone in China on RISC-V (or other open architectures). They are suggesting that for a US person to work with a person or institution in China, they must first seek a license from the Dept. of Commerce.

This would be like if in 2005 the US said that if you wanted to accept patches from China, for Linux, you needed their approval to do so.

This doesn’t make it good policy though. Our allies will be stuck in the crossfire and basically have to choose between Chinese RISC-V or US RISC-V. US export controls are very far reaching and apply to re-transfers, which requires a sustained administrative infrastructure to avoid violating. As a result European satellites and space hardware for example completely avoid using US parts to derisk compliance. What export licenses will ultimately do is isolate the US.

You seem to be arguing that the law is secretly a tactic to kill RISC-V, but I think you're giving the lawmakers too much credit.

Lawmakers can be a weapon itself rather than the holder of a weapon.

Banning RISC-V is like banning a programming language - more specifically something like banning a dialect of C or Python.

It's something that IP is composed of, not IP itself.

RISC-V is not a "chip architecture", it's just a standard instruction set. RISC-V International does not do "chip architectures", so open cooperation with RISC-V International doesn't actually help anyone design cutting-edge or "advanced" chips.

Yes, this is about 'RISC-V International' but at the same time they are sending an implicit message to western companies that they maybe shouldn't touch RISC-V if they know what's good for them (to some extent anyway).

The less western companies invest into the ecosystem the longer it will take Chinese to design cutting-edge or "advanced" chips (which I don't think is an entirely unreasonable view)

> The less western companies invest into the ecosystem the longer it will take Chinese to design cutting-edge or "advanced" chips

If by "ecosystem" you mean the actual RISC-V ISA this is quite wrong. The kind of IP that's actually involved in advanced chips has zilch to do with the ISA and is only minimally about the 'software-like' high-level microarchitecture; much of it is outright dark magic relating to the low-level physics of individual fabrication nodes. This is where China will always have the most trouble catching up with advanced, Western-aligned countries and where sanctions should be targeted. Anything else is just a pointless distraction and only hurts the West's ability to design cutting-edge chips.

I don't necessarily mean the ISA. But imagine if an US company designs a highly competitive RISC-V even if they don't directly license it would that help Chinese companies either directly or indirectly (e.g. people will port their software to RISC-V, major companies might adopt it etc. which would increase demand for other RISC-V chips and Chinese companies will unavoidably end up getting a slice of that pie with x86 and to lesser extent with ARM that's not a threat).

> This is where China will always have the most trouble catching up with advanced,

Yes I agree, but when/if they do they certainly don't want to be reliant on IP directly owned by western companies.

> if an US company designs a highly competitive RISC-V

Most of these will be proprietary. There is, however, a competitive open source chip design that the EU are putting money into for the European Processor Initiative, so good luck trying to control that. You can already go to github & download it[1]. (It has nothing to do with RVI in any case).

The software is already ported to RISC-V so that horse is running round the paddock already.

[1] https://github.com/openhwgroup/cva6

> Most of these will be proprietary

Yes, I'm talking about that situation more proprietary competitive -> higher demand/improved ecosystem for RISC-V chips -> Chinese companies become more competitive relative to AMD/ARM/Intel -> influence US has is diminished

> a competitive open source chip design that the EU are putting money into for the European Processor Initiative, so good luck trying to control that.

And for better or for worse than will benefit China.

> The software is already ported to RISC-V

I'm talking about software in general (Web servers, databases, mobile operating systems etc.)

> But imagine if an US company designs a highly competitive RISC-V even if they don't directly license it would that help Chinese companies either directly or indirectly (e.g. people will port their software to RISC-V, major companies might adopt it etc. which would increase demand for other RISC-V chips and Chinese companies will unavoidably end up getting a slice of that pie with x86 and to lesser extent with ARM that's not a threat).

You're really not going convince anybody who believes in the open-source ethos with that line of thinking.

Not directly transferring technology is one thing, but your reasoning is we should all refuse to use a common standard because if the US makes a really useful product based on that standard, then everybody worldwide would benefit from the positive effects to the ecosystem, and China would also benefit and that's bad?

The term "RISC-V" is usually used to include a collection of tools (Rocket, BOOM, etc.) But many (most?) of those tools are also open-source, so it's still pretty funny when people freak out about freely published information moving freely around the world.

China already has access to stuff more performant and efficient than rISC-V currently is… via both AMD and ARM. Why would this help at all?

Related https://news.ycombinator.com/item?id=38163412 is worth reading alongside this (it's not exactly a dupe, but it's pretty meta).

>"abusing RISC-V to get around U.S. dominance"

Saying the quiet part out loud.

> "a bipartisan group of US politicians is petitioning the White House to restrict transfer of know-how (so-called Intellectual Property, or IP) to RISC-V, as this may benefit adversaries like China."

No, it's because this may drastically cut Intel's and ARM's market share, and these politicians are dependent on such interests for campaign donations, jobs for their kids, etc.

I thought Bernstein v. United States settled that source code was protected speech?

Lol.

Court precedents aren't particularly reliable.

If you look at these chip Bans as being used to vault China into a leadership position they make more sense

1) China has home grown flagship tier chip

2) soon a monopoly on risc v development which they are already building around also protects Chinese arm investment

3) local manufactures soon won’t need asml

Perhaps the entire strategy is wrong

HN needs to get more sophisticated with its response to this risk. RISC-V is a lot more than AI; that's why it would be a tragedy if any restrictions happen. BUT there is a genuinely big emphasis on using it for AI tech. RISC-V summit is happening right now; there's basically an entire track of talks dedicated to matrix/vector optimization and proposals. We need to brainstorm acceptable compromises in case this continues picking up steam. Bad example but you get the idea: what if the restriction just got focused on RVV rather than RISC-V at large?

Its not like you can't change citizenship - and this is only a minor reason to do so.

Funded by Intel/ARM lobbyist I assume? A way to throw FUD at their competition?

Maybe: https://news.ycombinator.com/item?id=38166311

Meh. I think Intel has better things to do than worry about RISC-V.

Reply on news.ycombinator.com