Hacker News new | past | comments | ask | show | jobs | submit login
RISC V: A new blueprint for microprocessors challenges the industry’s giants (economist.com)
127 points by jkuria 7 days ago | hide | past | web | favorite | 94 comments

Gentle reminder: you can buy dual core RISC-V chips today for $9, FPU, DSP, and TPU included, with WiFi onboard if you buy a module: https://kendryte.com/. It won't run Linux (no MMU), but it will run FreeRTOS and TensorFlow Lite, straight from their GitHub.

There's uclinux, but I don't know how well it works (seems abandoned) and if there are other issues with it running on above mentioned RISC-V core.

I think it's part of mainline now (just like User Mode Linux/UML). So it's not abandoned.

Ah, thanks

It's not clear from the website that you can buy a module easily. Do you have to buy them in bulk?

You can buy them on Adafruit in the form of a device with battery, display and camera (but no wifi). Modules can be bought from Seeed or other usual Chinese places. No "bulk", you can even buy a single module. This, IMO, is one of the most exciting things I have seen in the embedded hardware space in years. At least all that Bitcoin money ended up producing something of value (the company that makes the chip stated out manufacturing bitcoin mining ASICs).


It won't run Linux (no MMU)

How are processes isolated from one another if there is no MMU? Is there some other form of memory protection? That doesn't seem likely?

Not having process isolation really limits the applicability of the chip.

It doesn't have processes.

This looks like to be a bare-metal (or RTOS) embedded system. It usually is for running a single application so isolation between processes doesn't make much sense. FreeRTOS doesn't even have concept of process neither, but tasks, which is similar to threads with preemption and deadline.

Furthermore MMU is often avoided in embedded systems because it violates hard real-time requirements: Translation table walking can take place while an interrupt occurs so it causes unpredictable worst-case execution time.

In case of ARM Cortex-M and R series they have Memory Protection Unit (MPU) instead of MMU to protect critical memory region.

Minix v1 and some older Unix clones run very well on 8086 and had processes. MMU make it easier, faster and more secure.

Will any BSD run on it?

The modern ones (FreeBSD at least) require an MMU. RetroBSD might, if they have RiscV support? Probably not, though. http://retrobsd.org/wiki/doku.php/start

(FreeBSD has RiscV support for versions with an MMU.)

For better or worse, this will become a basis that the semiconductor industry will heavily shift to homegrown chips in China. As problematic as Intel maybe, it still is a major source of cash, talent, and jobs. That will be eroded over time.

Unfortunately RISC V is decades away from getting anywhere close to Intel's x86 business, if ever. Currently it's nibbling away at the toes of the low end of the ARM market.

Even getting up to the level of competing with the mid ARM market will take huge investment. That is of course possible, but to do it RISC V will need to be backed by a very well funded body along the lines of the Linux Foundation, with strong support from big industry players. The sort of support that ARM already has. Until and unless that happens, I'm afraid it will stay a niche low end option.

RISC V is now affiliated with the Linux Foundation.


It only took ARM licensees about 5 years (2013-2018) to come up with competitive 64 bit chips like the (now sadly cancelled) Qualcomm Amberwing and Cavium ThunderX2. These chips are every bit as capable as Xeon. The decode stage is not a large part of any microarchitecture.

The PRC still largely does not manufacture or design semiconductors.

RISC-V is popular in that young industry because it is credible and meets their criteria, and that value is similar elsewhere (particularly in India).

> For better or worse...

If the world would let the PRC gain a lead in semiconductors, it would be an unequivocal moral failure; I say this with great personal warmth for Chinese people. The world is a more grotesque and unfree place with every gain of the CCP.

China has all the resources to lead in semiconductors. How can you stop this? Companies are being happily sold to Chinese investors all the time: Kuka, Putzmeister, Volvo, etc. Geely slowly takes over German carmakers: https://www.reuters.com/article/us-daimler-geely-electric/da...

Just wait couple decades...

> Just wait couple decades...

This is why I'm not just waiting, I'm getting richer, and I will continue not to do business or politics benefitting the CCP whenever possible.

That's all anyone can do. You can't help but lose if you give up before starting. If things go well, I can do more, maybe go help secure trade secrets in Taiwan, or document CCP controlled forign land and infrastructure purchases; if not, at least I don't see a coward in the mirror.

As quixotic as this all may sound, I genuinely appreciate this kind of candour about motivation. It's quite rare, more often to see these kinds of actions justified in terms of specious economic or technical arguments. It's as if people have become afraid of admitting that personal morality drives some of their decisions.

> It's as if people have become afraid of admitting that personal morality drives some of their decisions.

I think the fear is a bit justified. Admitting that you have moral principles means that people can hold you to them, and in fact that you will be held to them, as nobody is perfect. People with no moral principles can't be held to account, since they don't participate in morality! :- )

Yeah, it's just a pity though, because while individual efforts are laudable, they are insignificant unless coordinated as a part of a whole, so without this kind of honest public discourse it's hard to feel how we can get anywhere.

Are there western companies that do well after being bought by Chinese?

(Real question, because I'm only aware of the opposite).

Putzmeister (concrete pumps) is doing well. There is an article in German: https://m.dw.com/de/sany-und-putzmeister-eine-perfekte-ehe/a... I heard the same from locals living in the village with Putzmeister headquarters. Chinese investor let the company operate as it did before acquisition.

Kuka is doing not that good: https://m.dw.com/en/changes-at-german-robotics-firm-kuka-rai... But I am not sure if Chinese investor is the problem. Kuka had often problems in the past.

It’s not enough statistical data for serious conclusions, but I assume that these investors are trying to buy more companies in foreign countries and not to scare locals.

Thanks. I heard about Kuka having problems and wondered if that applies to other now-Chinese companies as well.

> The PRC still largely does not manufacture or design semiconductors.

This is thoroughly wrong. China has a large number of fabless design places which have been going for years, and is now starting to get into manufacturing in a big way. Lately it's been driven by the trade war (so that worked well ...)

Coreteks did a really nice video on Risc V.


RISC-V is, among other things, a single shining ray of hope that we might actually be able to build a truly secure computing platform from the ground up instead of the backdoor-riddled stacks we have today.

Maybe, but I don't see why vendors making RISC-V chips are inherently more trustworthy? There are lots of dodgy vendors building standardized parts. It seems like they'd be easy to counterfeit?

We might hope that more competition will mean it will be harder to install backdoors but other than that there is no guarantees.

Assuming that USA, Russia, and China are not hidden allies then we might be able to buy risc-v manufactured chips from other countries to prevent domestic spy agencies from accessing private data, but then there is no guarantee that USA, Russia, and China are not allies. Perhaps they share data with each other at similar capabilities to Five Eyes.

At the worst case we got some open source chip design that will constantly evolve. Perhaps we might be able to print such CPU chips at home but that is decades away.

It does allow you to compare a chips behavior to what you'd expect from a public design. That's not something that just anyone can do. And if the design has a stage with a few FO4s of latency free and some unused silicon in the right place they could still put in a secret feature without anyone noticing. But it's got to be a pretty simple secret feature for it not to be obvious what's going on.

Could you use side-channel power analysis to ensure a chip is on-spec?

No, the backdoor can use some activation mechanism on chip to avoid drawing power when not in use, then the additional power draw will be so low that it's essentially undetectable.

Tbh you'd have a hard time since either you're trying to measure a few milliwatts in a very noisy environment (either very low wattage, so various effects dominate, or high wattage, where your equipment will quickly get very very expensive).

I've worked with the chip-whisperer in the past, it's definitely a noisy environment.

For sure! I'm really looking forward to seeing production runs of https://keystone-enclave.org/.

Consider joining and supporting the RISC-V Foundation! https://riscv.org

I don't buy the argument that having a bunch of vendors extend the ISA makes it more secure. I think we would end up with an ecosystem with a lot of custom extensions and functionality that will end up exposing more attack surfaces and that would make it difficult to know if your device is affected or not.

You are highlighting the risk of fragmentation causing security problems. I think that's valid.

I think a lot of people are talking about "closed source" security as in nation-state level actors and "backdoors": https://www.csoonline.com/article/3220476/researchers-say-no...

"Alibaba, an e-commerce giant based in Hangzhou, announced its first RISC-V chip in July"

What did make RISC-V a game changer? Why did not any open source chip enjoy such a success before?

Timing (trade wars, ending days of Moore's law, intels incompetence as of late), marketing (ucb/Stanford), few outright dumb decisions (e.g. no branch delay slots), money (DARPA), industrial support (nvidia, western digital, esperanto etc.)

Moore’s law is pretty well on track – transistor density has been steadily increasing for decades now, and still is.

Things are still shrinking but the tempo seems to be slowing down a bit even at TSMC. Maybe we'll get our full 6 remaining doublings down to 9Å[1] but I'm not sure I would count on it, especially given that the cost of a cutting edge fab is also going up exponentially. Even if features stop shrinking that doesn't mean that price per transistor will stop going down. And we're very far from the theoretical limits of energy per computation so I tend to expect we'll move to some other computational substrate in the future.



1) Slowing is happening and at some point you run out of atoms

2) Usually people mean all of CMOS scaling (ie Dennard Scaling) when they refer to "Moore's law ". Exponential increase in power density is low key useless...

There are a fair number of folks including Gordon Moore who have said Moore's law will end in the next couple halvings.

Why is no branch delay slots a bad thing?

You read that wrong. GP said branch delay slots are a dumb thing, and the RISC V doesn't have it.

More details on Wikipedia[1], and there's some nice answers as to why it's a bad idea these days on SO[2]. To summarize, the branch delay slot optimization makes exception handling and debugging much more complicated, doesn't bring much added performance if you have a deep pipeline, and modern branch prediction can render the the branch delay slot optimization moot anyway.

It should be said that it wasn't necessarily a dumb thing back in the days, but rather that it's a dumb thing on a modern architecture.

[1]: https://en.wikipedia.org/wiki/Delay_slot#Branch_delay_slots

[2]: https://stackoverflow.com/questions/54724410/why-is-the-bran...

Probably should have been:

  Why are branch delay slots a bad thing?
Branch delay slots were used in early pipelined RISC processors: MIPS, PA-RISC, SPARC. These were short pipelines but not superscalar and not out of order superscalar. Navigating pipeline hazards gets much more complicated with much deeper pipelines, more functional units and out of order execution. Basically, branch delay slots don't scale.

It came out of Berkeley, and David Patterson contributed to it. That gave it a lot of initial clout. Darpa gave them early funding as well. And to some extent, I think the timing was just right vs people's frustrations over ARM.


A lot of small things added up

A strong team

I think democratized fabless is a mainstream thing.

I’m surprised the Economist doesn’t know ARM is British!

It isn't anymore - it became Japanese a couple of years ago.

Chris, would you still consider, Jaguar, Bentley, Land Rover British?

Because I think ownerships changes with time, but most of these brand, thinking, management are still very much British.

But this isn't an article about corporate culture - it's an article about the nationality of the company which owns the IP.

> are built around designs, known as instruction-set architectures (isas), which are owned either by Intel, an American giant, or by Arm, a Japanese one

Who owns the ARM IP? A Japanese company. Not a British one.

If you asked me if ARM has a British history and possibly culture, then yes (it certainly pays like a British company!) But that wasn't the context - the context was who effectively owns the IP.

So what do you think is more likely? The Economist are mistaken about the history of the company, or they were using a different context, the one they actually make explicit in the article?

I was just wondering from a British Point of view whether you still consider them to be British or not.

>Who owns the ARM IP? A Japanese company. Not a British one.

ARM, a British Company, owned by Japanese Funds.

Because to me that is like saying HSBC is owned by Chinese if HSBC had Ping An and other shadow investors owning majority of it, but certainly HSBC is a British Company right?

P.S - I means no offence in case any Brits got the wrong end of the stick. Just genuinely curious.

> I was just wondering from a British Point of view whether you still consider them to be British or not.

I think they're culturally British. I don't think they are British owned or that their IP is British owned. Which was the context of the discussion.

Mostly no. They may be based in the UK, but the decisions of whether to introduce a new range, hiring, firing and where profits go lie elsewhere.

The Land Rover office may still look and feel British, but it's Tata gets the final say, and perhaps one day decides they will all be made in India. Or perhaps Poland post no-deal Brexit.

JLR are already moving a lot of production to Slovenia and China, FWIW.

No. The MINI isn't British either (they're great cars though).

Bought by softbank

With major design centers in Austin (A76) and Sophia-Antipolis (A53) as well as Cambridge.

I wonder where's AMD... did they omit it?

AMD aren’t part of the RISC-V foundation.

Anyone hiring an ARM refugee willing to work on RISC-V?

I too having been asking around. Did not have much luck. Most of the requirements are in chip design. Software contribution is done by the open source community. Corporations are contributing to open source but the team size is very small.

Most open-source software is written by people being paid to do it, and at SiFive we are expanding the software team as well as the hardware side. Ex-ARM people are certainly welcome and may recognize a few people when they get to the office(s) :-)

Have you tried SiFive?

I have an off-topic rant: I know it's "RISC FIVE" and not "RISC VEE" but as long as companies use letters to represent numbers it's my opinion they forfeit the right to complain that people aren't saying it right. Apple finally gave up and changed Mac OS "ex" to macOS; everyone else should follow suit.

interestingly enough, the “v” in risc-v also means “v” as in vector [1]

> and so we named it RISC-V. As one of our goals in defining RISC-V was to support research in data-parallel architectures, the Roman numeral ‘V’ also conveniently served as an acronymic pun for “Vector.”

[1] https://people.eecs.berkeley.edu/~krste/papers/EECS-2016-1.p...


so maybe in a way “vee” isn’t so inappropriate after all ^_^

The roman numbers were not really used for RISC 3 and 4, (SOAR and SPUR), only for RISC I and RISC II whose development overlapped.

But then went on to call the iPhone 20 "iPhone X". And then switched back to Arabic numerals with the iPhone 11.

Not to mention the "Plus" vs "Max" and all the other naming convention mismatches.

Anyway, I agree with you.

iPhone XR is the worst. The "X" is a number, the "R" is a letter. I just can't see it as anything but "ex-are".

pretty sure the only people that call it ten-are are apple store employees

and that's only because tim cook will haunt their dreams if they don't

Some people persist in calling "GIF", "jif", so whaddayagonnado?

I pronounce it with the hard "g", and I'd happily ignore the few people that pronounce it with the soft "g", except that the guy who invented the format intended it to be pronounced "jiff".

Yes. Being old, I remember. Well, considering that the "G" stood for "Graphics" and that "Graphics", in my native American English has a hard-g, I'm just going to declare the author of the format to be a troglodyte and go on from there. (Not that there's anything wrong with that - many of us tech-types are troglodytes and misanthropes and lack a classical education)

I insist on the pronunciation "jayfeg" for JPEG, for the same reason.

Do you also pronounce 'giraffe' and 'giant' with a hard "g"?

Thers's an idea - revise the expansion of the acronym to say Giraffics Interchange Format. Make it an ISO standard and donate proceeds of sales of the paper version to protecting the giraffes. Problem solved.

jrafics interchange format is exactly how you say it long. Why is this even a discussion?

“joint potographic experts group”

Do you pronounce "gift" with a soft 'g'?

No, but "gift" is a borrowing from Old Norse. "GIF", as far as I know, is not an Old Norse lexical item.

It's still a very common English word that's spelled very similarly. One can hardly be surprised if people guessing at the pronunciation of "GIF" select a hard 'g'.

Sure. So English has some words which start with "gi-" which have hard 'g's ('gift', 'give', 'girl', 'giggle') and some which have soft 'g's ('giant', 'giraffe', 'gibber', 'ginger'), so both pronunciations are actually reasonable in English for the orthographic form 'GIF'.

I'm glad we agree.

I always say "risk-vee", and everybody knows what I mean.

I also say "macosix" and "windos" (and used to say "solarix") as generic alternatives to commercial trademarks.

Marketing departments can only propose pronunciation, they cannot dictate it, and they have no right or power to prevent inventing synonyms for the trademarks they control. They also have no authority over verbs, so e.g. "googling" is a generic term right out of the gate.

There won't be a RISC-VI, so it doesn't matter that it means "5" to some people.

I'm not sure it would be less confusing to name the ISA 'RISC' instead of 'RISC-V'. The 'R' in 'ARM' stands for 'RISC'.

RISC is a generic term for ISAs with few instructions (in comparison to CISC ISAs).

Increment the letter V by three positions and call it RISC-Y.


Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact