Hardware, I guess? ARM's moat is that you can buy a fairly RISC CPU today for a fairly cheap price. It won't stay that way forever though, and there are efforts underway with RISC-V to drain that moat.
The interesting part (to me) is speculating which parts of the market ARM will choose to focus on. Microcontrollers seem like a lost cause at their price point, but making a shift to higher-margin licensing might be part of a strategy to acknowledge that loss.
Both Google and Apple have multi-arch support and Apple has pulled off a ISA transitions, guess that really only leaves Apple's investment in the actual M1/M2 chip-design itself.
Apple's price-point and margins are high enough but they're pretty protective of their margins afaik. I wonder if it's worth transitioning away though.
I think this will definitely boost Risc-V investment and maybe other alternatives, if for no other reason than to have a viable threat to leave ARM if they decide to get more aggressive in pricing.
Apple had a secret x86 port of MacOS System 7 in the early 1990s ("Project Star Trek"). Similarly, Apple continued to maintain a public x86 port of Darwin and a secret port of Mac OS X ("Project Marklar"), prior to Apple's transition from PowerPC to Intel. They no doubt have a secret RISC-V port of macOS today.
> They no doubt have a secret RISC-V port of macOS today.
Possibly but as it’s believed they were lead partner for the design of the ARM64 ISA and seem to be able to add their own extensions (as well as designing their own micro architecture) they already have a lot of control in a way they didn’t with 68k and PowerPC.
I doubt Apple still maintains macOS for PowerPC. Building for RISC-V would be to keep future options available, whereas PowerPC is a dead end.
We might look for clues in any Apple contributions to LLVM and Darwin for RISC-V or PowerPC.
I went searching if Darwin still supports PowerPC. I found a Reddit thread about compiling Rust code for Darwin PowerPC, but they only talk about old versions of Darwin:
I really doubt they have seriously considered porting macOS to RISC-v though.
It shouldn’t take that long to port darwin to a new arch anyway. And when/IF RISC-V actually becomes even close to being competitive to ARM/x86 it will probably be very different than it is now.
> Microcontrollers seem like a lost cause at their price point
Is there even a RISC-V that can compete against the ARM Cortex-M33? Or ARM Cortex-M0+ ?
ARM has the microcontroller market in a deathgrip. I'm honestly not seeing any competition here from anybody. STM32, NXP, Microchip SAM, etc. etc. Its all ARM.
You get some 8-bitters (8051 lol still alive) at the lowest end with SiLabs USB stuff and other such stuff. But ARM is incredibly dominant here.
Not yet, most of the RISC-V cores intended to compete with those chips are still on-paper and not in-production. Tables turn though, and the messaging ARM is sending right now seems to be directed away from their smaller piecemeal component market. That makes sense, too - Chinese manufacturers have been mass-producing knockoff ARM processors for years. There are probably more unique STM32 clones than there are unique ARM core designs available today.
The competition isn't here yet, but the bloody strangler is knocking at the door and someone upstairs just yelled "Nobody's home!"
Robotics and IoT edge is another. The bl808 is a 5 core chip for $2odd with the lowest power mcu having access to 64mb ram, no other mcu can do that as far as I know:
In the low and mid end there's plenty of competition. Andes, SiFive, Cortus, BA Semi and probably more. Many of them used to offer their own proprietary ISA and have moved to RISC V to leverage the ecosystem (software and tools). Instead of a semi frozen old GCC version their customers can now use a recent compiler directly from upstream for example.
Renesas has started replacing ARM core by RISC V in their new micro-controllers. They still offer their ARM products and will do for a long time, but move new ones to RISC V.
ARM's advantage is at the high end. For a high performance core it's key to work closely with TSMC and Samsung on the most advanced nodes, to tune the design to what leading edge processes can do. This is very labor and capital intensive. It definitely gives ARM an advantage against smaller IP providers at the high end. But at the low and mid end? Nope. It's where ARM is easier to replace.
> In the low and mid end there's plenty of competition
None of these chips you're talking about are on Digikey or Mouser though. I dunno, are they not selling these chips to the big retailers in order to try to cut costs down or something?
I'm not talking about chip makers, but about CPU IP providers. Like ARM.
If you say there are not many publicly available products including publicly accessible RISC V cores from those companies I listed compared to ARM, yes this is true. Because they're (for now) mostly used for embedded application, deeply inside complex SoC. Only used by the chip makers developers.
A relatively small company like Andes has their IP in over 10 billions SoC today. A lot of it for things like touch screen controllers, but also more complex things now. But I'm not sure you can find a mass product with an accessible RISC V Andes core.
Still, public access has already happened with Renesas, and it will likely get more common. People tend to focus on the high end, but the most common in high tech is to attack from below. Take a foothold in the low end because you're cheaper (smaller companies with lower costs can deal with lower prices), build a customer base, then grow from this. Until one day you can attack the high end (we're not there yet IMHO). I have access to these companies roadmap and this is what I see them doing, and it shouldn't be surprising.
But this means that at this stage, the majority of RISC V use is invisible to the public. To me this is perfectly normal. Building an ecosystem and maturing a technology like this takes a long time.
IMHO we're already at a stage where ARM has real competition in the low and mid end, particularly where there's little dependency to the ISA (embedded, or simple uC programmed in C anyway with no 3rd party binaries included, like Renesas). I think ARM is still safe at the high end though, and it will take some time for RISC V to get there. Chip making is hard, particularly at the high end.
"Is there even a RISC-V that can compete against the ARM Cortex-M33? Or ARM Cortex-M0+ ?"
Yes, of course, lots of them. That is where RISC-V has taken the most market share from ARM so far.
SiFive's E20, E21, E24 hit various points of that. Plus they offer the S21 which is a 64 bit microcontroller in Cortex-M33 class. ARM doesn't offer anything like that -- if you want 64 bits from ARM then you have to take the full no-subsets ARMv8-A ISA.
The Europeans have their Ibex (formerly Zero-RISCy). Andes has the N22.
What's the ULPMark (power benchmark) on any of those SiFive RISC-V chips? STM32U5 (one of the many ARM Cortex-M33 designs) is 464 ULPMark-CP.
I don't think there's even a power benchmark on any of those chips you've mentioned. At least, I'm not seeing any online in my searches.
STM32U5 (and most other Cortex-M33 based microcontrollers) are ~110nA of power consumption sleep mode, and 16uA / MHz at 160MHz, some of the most power-competitive chips I'm aware of on the market. I'm talking "run a microcontroller on a CR2032 cell for 10 years" kinda thing here.
--------
EDIT: Looking at the ULPMark website (https://www.eembc.org/ulpmark/ulp-cp/scores.php), it looks like the #1 chip is RSL15, which is ARM Cortex-M33, as I expected. With a ULPMark score of 1090 (higher the score, the less power the device uses). Nearly everything on the top list is Cortex-M33...
Except LP5100 and XRM32UL051 are some Chinese Chips I don't understand. They got some impressive scores but I don't know what architecture they have. Nearly everything else is Cortex-M33 (except for the Ambiq, which is ARM Cortex-M4).
Cortex-M0+ was the king of this a few years ago, but M33 has displaced it. We still see Cortex-M0+ chips due to much lower costs around though. So if RISC-V is around Cortex-M0+ level, I'd still be interested.
"What's the ULPMark (power benchmark) on any of those SiFive RISC-V chips?"
SiFive, like ARM, doesn't make chips, they license cores to people who make chips. The power characteristics are up to the chip designer -- what node they use, which "corner" of that node they use, and various other chip design features -- and not the core designer.
I don't know of anyone who is putting out stand-alone microcontroller chips using SiFive cores. They've been concentrating on getting them into larger systems. For example Qualcomm said a few months ago they've shipped 650 million SoCs (i.e. Snapdragon XXX) with RISC-V cores in them (probably SiFive as they're a SiFive investor).
The closest I know of (using SiFive cores) would be the Bouffalo Lab chips, but they are not just microcontrollers but WIFI/BlueTooth which of course by their very nature use considerable power. The BL602 datasheet says it uses 500 nA in "hibernate" (RTC or GPIO wakeup) and the CPU 22 mA at 192 MHz with radios off, so 115 µA/MHz. Certainly not in the league you quote.
Well, actually ... I just looked and the STM32U5 datasheet says 440 nA "standby" with RTC or GPIO wakeup, so actually 440 vs 500 is not all that different, comparing apples to apples.
> ARM has the microcontroller market in a deathgrip.
No they don't actually. RISC-V is already selling billions of devices and its growing at an absurd rate.
You have companies like WesternDigital transforming all their products to RISC-V. And a lot of harddrive and SSD vendors are following. You have companies like Andes that push RISC-V onto every cheap consumer device from China. Companies like SiFive are pushing into the automotive market. NXP has RISC-V products already. Companies like Gaisler into the space market.
RISC-V is in more places then people realize and its still very early in the adoption curve. There are like 100s RISC-V companies pushing RISC-V into every niche.
> RISC-V that can compete against the ARM Cortex-M33? Or ARM Cortex-M0+
In terms of performance, RISC-V offerings are already beating them. ARM has momentum but in terms of performance per area RISC-V is better.
I seems I might have been mistaken about NXP, they are part of RISC-V foundation and have worked on some parts where the core was an OpenSource RISC-V core. But I couldn't find an actual product.
As for the Cortex-M33, I don't know if RISC-V cores in your link. So not sure we have a great bases for comparison.
If if it is better now, there are so many more companies competing on core design, I don't think ARM can consistently outperform all of them in all segments. RISC-V is just a superior business model and it will eventually just win.
I mean, 8051 isn't dead yet. SiLabs continues to make extremely competitive UniversalBee 8051 designs.
In my experience, bet on legacy. There will always be enough time to wait and switch later if things turn out better.
This is doubly so in embedded, where legacy issues / manufacturing issues reign king over performance.
-------
Cores are often the wrong thing to focus on anyway at the embedded level. I don't think performance is a major concern. Maybe power/performance is a concern in battery constrained apps, but many electronics will bottleneck elsewhere.
So I'm really not sure how or why RISC-V would break into the uC world / low end.
If anywhere, they have a chance at the higher level, like at Rasp Pi level chips or stronger.
Of course these things never die, MIPS wont die. ARM wont die. Nobody is claiming they will. But also, you don't start very many a huge amount of new project with 8052s.
Performance isn't that important maybe (depends on the application), but energy often is, size is, having lots of vendors is, having a large software base is.
> So I'm really not sure how or why RISC-V would break into the uC world / low end.
If by lower end you mean 8-bit, then no it wont. But one of the large adopters of RISC-V are actually places where there used to be tiny 8-bit cores and now people want more substantial 32-bit cores and RISC-V is really good there.
And it has already broken into that market as there are already lots of RISC-V products shipping, lots of them in lots of consumer electronics from China.
RISC-V will continue to grow in that market, if you were a MIPS costumer for example, you are likely gone move on to RISC-V?
> If anywhere, they have a chance at the higher level, like at Rasp Pi level chips or stronger.
At the very low end being very cheap, not involving as many lawyers actually matters quite a bit because volumes are high. Specially when there are very competitive verified opensource cores as well.
Not to mention all the commercial offerings (or commercial offerings that support those open-cores). They might not be open, but its easy to get up and running with RISC-V compared to commercial stuff.
There is just so much education moving to RISC-V. Lots of people will be educated in RISC-V and lots of people will look at some of these OpenSource cores to learn. When places like Berkley, Stanford and ETH embrace something it usually trickles down to most universities.
I don't think they are talking about the microcontroller (Cortex-M) business, this is about ARMv8/9 in smartphones, tablets, computers. What the two have in common is the name "ARM".
Designing a high-performance CPU within the power budget of a smartphone is expensive and hard; so hard that Samsung, Qualcomm, NVIDIA, and even Intel have all tried and bowed out because ARM did it better. RISC-V doesn't really make this easier, since at best the ISA is neutral at the performance levels of the Cortex-X cores.
Outside of application cores though, no real moat.
Well depends, Apple has been refining their arm chips for many generations now. Generally the Arm competition (non-apple) hasn't caught up to Apple yet.
CPUs are amazingly complicated. A major new CPU designs can take 3-5 year, consume many $100M, and end up not competitive. Even Intel has had serious issues after many years dominating their market.
The ecosystem of OS kernels, platforms, testing/verification tools, compilers, debuggers, and related can take awhile to mature. Android is a major platform for phones, and while by default apps are portable, assuming the vm is ported, but some apps write native code to avoid the Java related overhead. Intel tried a phone and ended up emulating ARM chips for native arm code and delivered terrible performance and terrible perf/watt. That project died with a total design win of (I believe) just a single relatively low volume phone.
So making a CPU to compete with a flagship android phone isn't easy, and even if investing some multiple of $100M and years you might end up still behind, or bankrupt when you scare Arm into cutting prices again.
> Arm wants to stop charging chip vendors to make Arm chips, and instead wants to charge device makers—especially smartphone manufacturers—a fee based on the overall price of the final product.
Device maker: I'd like to buy some ARMs please.
Arm: for what final product?
Device maker: none of your business.
Arm: we're making it our business. Tell or get lost.
Device maker: if you must know, we put them in our Mars rocket...
According to the report, Arm wants to stop charging chip vendors to make Arm chips, and instead wants to charge device makers—especially smartphone manufacturers—a fee based on the overall price of the final product.
Hmm, I wonder if this could put the brakes on the vomitously expensive phones that dominate the market these days.
I recall news about Qualcomm charging a fee based on the overall price of the final product too. But does Arm have as strong of a hold on the market, compared to Qualcomm, to pull it off?
The original Arm proposition was to make it cheaper for firms to use Arm’s IP than to develop their own.
I suspect we are now in a (possibly short) period when there is no viable alternative to Arm on mobile say.
When that period ends then the alternatives (RISC-V based) won’t be free. There will still be a cost.
This move obviously looks like a move to juice profits short term ahead of a float, but perhaps also to tilt pricing to keep Arm competitive on the lowest end devices.
There's this RISC-V dream about an high performance CPU with a fleshed out, open source ISA, with a company championing it with a big proven core you can buy, and decent compiler support in place... but we already have it! Its right over there!
Also, I think POWER is a bad omen. If IBM can't sustain POWER's hype (even with good CPUs you can buy right now), I'm not sure how RISC-V is supposed to surpass that from square one.
I get the impression that devs are happy about RISCV and willing to port things to it, not so much for POWER. Also China seems to be pursuing RISCV heavily.
But where the rubber hits the road seems to be in niche stuff like accelerators (which is indeed a great application for RISC-V) and low power embedded stuff.
Other countries (like Russia, and China's MIPs thing) have pursued novel architectures for higher power general purpose computing, but its mostly for government use I think. China pulling off a societal transition to RISC-V would be quite a coup.
To be fair, a lot of software already supports POWER right now, including the Linux kernel, compiler toolchains, the OpenJDK, various database engines, and so on. Don't forget that a lot of embedded devices used PowerPC in the past and it was easy to get PowerPC based hardware like e.g. old Mac computers or game consoles. Today you can get large POWER based servers which compete with big multisocket x86 servers.
I think the main problem of POWER today is that you can't get a "Raspi sized" development board, while you can easily get all kinds of ARM based SBCs as well as various RISC-V based SBCs. The smallest POWER system you can get right now is iirc a workstation from Raptor which essentially competes with Xeon workstations, if you don't want to play around with soft cores on FPGAs.
Is there any actual POWER core ip out there that people can build on or experiment with? Yes, the ISA itself is open, but last I looked that was it. I think RV's availability of implemented cores (even if they aren't the most cutting edge or high performance) is helping it build some momentum. Whether that's actually "ARM-killer" momentum remains to be seen, and might well not be... but I don't think I'll write them off quite yet, just based on IBM's long history of failures to position and market POWER well.
> I just don't see a line from here to where ARM is now
I don't understand why. RISC-V is being used and pushed by 1000s of companies at this point. You can license cores from 100+ vendors or you can make your own.
There are lots of companies pushing it into every single part of the market. There are well kown cores for space travel, the next generation is RISC-V. RISC-V is already in plenty of low level cores, in things like SSDs, NICs and so on. Companies like Andes push it into billions of consumer devices.
In terms of the ISA itself it is already a match or better then ARM in most ways. And in 1-2 years with the next rounds of specs, it will be more advanced.
To me it seems like RISC-V is already winning by far and its just a question of time. This is more like Linux in the late 90s vs commercial Unix.
1990s sounds seem to be all the rage in pop music right now. But the only ‘90s revival I want is the POWER ISA. Sadly it seems more unlikely than Sinead O’Connor appearing in a Justin Bieber track.
POWER was fake open, while not actually being close to open for a long time. When RISC-V gained adoption they made a panic move and opened it up.
But there is basically no reason to pick POWER over RISC-V at this stage. Not unless you are already invested in the POWER architecture.
RISC-V has far more OpenSource work, far more companies involved, the spec is driven and improved by a lot of people and so on.
The reality is most work on RISC-V doesn't go into super high performance stuff and if its there, then usually along side high performance AI stuff. SiFive, Esperanto,
Tenstorrent and so on.
> I'm not sure how RISC-V is supposed to surpass that from square one.
Softbank owns ARM but is planning an IPO this year. After Softbank reported massive losses last year, its legendary founder Masayoshi Son said he was stepping down from Softbank operations to focus on “unleashing ARM’s potential”. This seems to be part of his profit drive.
ARM is an incredibly important foundational company but they barely make any money, hell they lose money about half the time, including during the pandemic, despite absolutely crazy electronics sales!
Softbank isn't a charity, they want $40b+ valuation on a company that makes $0 +/- 1.5b ish. They want almost double what they paid for it, in fact.
The only companies willing to pay that much money are the ones who see some kind of strategic synergy with owning ARM. NVIDIA could have gotten CUDA into every smartphone's default graphics IP, and become kingmaker licensing one of the most important ISAs in the world. Not only would they not kill that "just to sell some more tegras", but actually it likely would have been a watershed moment of NVIDIA moving towards licensing their IP to external companies. This would have been the start of the pivot towards semicustom - just like AMD already does.
People just don't want to accept that because of "green man bad syndrome". But NVIDIA isn't exactly what people always present them as either - they are moving towards open linux drivers, they opened up to adaptive sync, both of which people would have (and did) defend to the death would absolutely never happen, hell would freeze over. And now that they are, still, nothing is ever good enough - well they have a closed userland, they have blobs (which so does AMD, in linux-firmware), they have a security processor! Even if those things are also true of other brands... the green man bad field means it'll never be good enough.
There were always two alternatives to NVIDIA buying it - another sugar-daddy with business synergy, someone like TSMC ("bundle deal if you get it fabbed here") or Samsung or someone like that that could pair it with some other aspect of their business. Or finally, Softbank just makes it actually turn a profit. Who pays for that? Well, big tech, and probably ultimately consumers too.
Companies like Google and Amazon and Tesla have grown fat on the "pay almost nothing for this ISA, bolt on our accelerator, sell it for billions" while ARM gets nothing and that's never been sustainable in the long term.
Companies like HDMI already do make private tech that is licensed per-device, and again, you can't even have HDMI 2.1 in linux on AMD because it's closed-source and they won't license it in a compatible way, while NVIDIA can offer support because of those blobs. Qualcomm requires similar per-device licensing on their modems too, and wireless/modem blobs are some of the most notoriously opaque/blobby peripherals too. People deal with it, pay the licenses, and move on.
And the "big tech forms a consortium and buys it and runs it as a nonprofit" was always a pipedream. Never going to happen, and none of the companies involved have any incentive to push tech forward in the same way as, say, Apple does. They just need it to be Good Enough And Cheap, and it already is. And scraping up $40b to keep it cheap is a bit of a mixed bag.
Well, you didn't get NVIDIA. You did get big ARM layoffs (just as Jensen warned, and just as people scoffed at, ARM UK dropped 20% of their staff) and big price increases. Just as predicted. /shrug
And yeah RISC-V exists and it's good to have an alternative to x86 and ARM too. There's a bit of a lifecycle here where they tend to get stagnant over time, maybe RISC-V being open-source will help that. But it also is going to be hell for ISA compatibility across vendors (a lot of the codes ar...
I agree with your business analysis, though I take no position on "green man bad."
I'd add to your analysis that, given the immediate IPO-driven need to demonstrate they can increase margins, raising prices sooner is probably better than later because RISC-V is not yet a viable competitive alternative for ARM's most attractive customers. However, later it very likely will be and if RISC-V proves to be a threat to ARM's blended margins, ARM can drop prices then because at least the IPO will be over.
It's worth noting that this is the argument Qualcomm uses for its "FRAND" licensing terms on its modem patents.
I did like 'Arm has been "frustrated by customers’ reluctance to accept the new arrangement."' why ever would that be? why would customers not love this value proposition?
ARM is so foundational, it should be bought by 3+ companies like Apple/Qualcomm/nvidia and turned into essentially a Chip ISA Lab where basically all of the companies that want to license stuff can pretty much do whatever they want and pay no licenses other than some set of dues that each company must pay each year to keep the thing alive. Then softbank gets its annoying payday for its dumb investment, and things can go back to how they were.
So, funny story, that's actually how ARM started. Apple wanted to use the Acorn Archimedes chip - the Acorn RISC Machine (ARM) - in the Newton. Acorn spun off the chip design part of the business and got VLSI in on the deal, in a business alliance structure very similar to the much more famous AIM (Apple, IBM, Motorola) alliance for PowerPC a year later.
The problem here is that ARM's current owners - SoftBank - are starving and trying to eat their seed corn. The problem with a company like ARM is that's value is mainly that they don't take all the value of the widget. This is in contrast to, say, Intel or Nvidia who completely squeezed all the margin out of the PC business decades ago. Problem is, "we leave money on the table for our customers" is not something the stock market will buy, so they have to start squeezing.
(Same reason why Nvidia tried to buy them.)
Personally I don't see this being fixed unless ARM is either privately owned by someone who just has the money to burn on an FRAND-licensable architecture, or they get nationalized.
Seems like an appropriate point to give credit to Robin Saxby who, with help from Acorn and Apple - notably the late Larry Tesler, devised the strategy and who then energetically made it work. Saxby is a great, engaging, interviewee.
First of all, that is not how ARM started, because at that point "ARM" already existed :)
The collaboration with Apple failed spectacularly. The only reason ARM is alive today is because Nokia and Ericsson needed a better CPU for their phones and ARM happened to be really power efficient (mainly by accident, but that's a story for another day).
That's a slightly roundabout way of saying that Arm built the best microprocessor design for the fast growing mobile market. A design that was a factor in Nokia getting to No. 1 in the handset market.
The 'happened' on power consumption is true on day 1 in 1985, but the Arm team continued to keep power consumption low and added features, such as Thumb, that won Nokia and others over.
Shameless self plug for more on this stage of the Arm story:
ARM is a mess as an architecture ( every ARM has different ISA, you need blobs for the majority of it to make it work and OS support is lacking). RISC-V is almost nonexistent (where can i buy one at a reasonable price? ) and it seems that it has the same fragmentation problem.
In fact, one of the main reasons for RISC-V's existence is fragmentation - the ISA is meant to be extended and not meant to be one-size-fits-all. The core of the ISA is designed for everything from microcontrollers to supercomputers, but extensions tend to fall closer to one side of that.
Or, an other way to put it is, it’s an attempt to build diversity without fragmentation. The spec follows the open closed principle in a sense - it’s open for extension but closed for modification, and if you choose to target say RV64G, you know that it will run on most 64 bit RISC-V processors. If you choose an extension, it’s not all that different from, say, figuring out which of Intel’s various SSE extensions is actually available.
Of course, there are vendors who are only RISC-V adjacent, but they will either put up, shut up or subsume the entire ecosystem if they get big enough. Time will tell.
ARM has two main architectures right now: 32-bit v7 and 32/64-bit v8-9. They sometimes run compressed ISA (eg thumb2) but that is still part of thr same architecture. I don't know what the blobs you mention have to do with this
RISC-V despite its young age is even more fragmented. This is because the Berkeley folks took their sweet time to finish important parts of the spec and in in the mean time multiple companies shiped out RISCV with home grown ISA changes.
There was no realistic chance of Berkeley to 'finish' the spec. The specs were finished in the foundation not just by Berkeley.
Fragmentation is a feature of RISC-V, and often it represents actual diversity in use, more then just needless fragmentation. And the majority of the time its additions, not changes to anything that in the spec. And RISC-V specifies how that it is be done in order to stay compatible.
In terms of software for Linux its not actually fragmented. Pretty much all major linux and free software distribution agree on targeting the RV-A profiles.
When was the base instruction set published? When was the privileged ISA finalised? How many CPUs with a home grown memory protection designs were sold between those two dates?
2015 and July 2019, though the HiFive Unleashed shown in late 2017 and shipped in February or March 2018 ended up being compatible with the privileged ISA ratified in 2019.
> How many CPUs with a home grown memory protection designs were sold between those two dates?
Zero that I'm aware of. Want to inform us?
The Canaan Kendryte K210 dual core 400 MHz 64 bit microcontroller shipped with an incompatible MMU, but it wasn't home grown — it's unmodified Berkeley Rocket implementing the current version (Priv 1.9.1) as at the time Canaan took the core snapshot for their SoC.
If that chip was a serious Linux target then priv 1.9.1 support could be put BACK in the Linux kernel (it wasn't upstreamed but it existed and was published). But with only 8 MB of non-expandable on chip SRAM there is no point anyway.
Over 10 billion shipped so far. Chances are you're using a few without realising it. Got an Android phone made since 2019? Qualcomm has shipped 650 million Snapdragon SoCs with RISC-V in them. Got a WD hard drive? Got an NVidia GPU made in the last five years or so?
Those are the extremes of the mainstream range you can buy right now. Considerably higher performance is rolling out all year, including many-core (e.g. 64) OoO server chips.
This adds context to the Qualcomm/ARM spat. ARM desperately wants to find anything to hold Nuvia hostage so that Qualcomm can be levered into the new charge-for-value licensing model. Before I was on the fence about the lawsuit - there were arguments on both sides that made sense. Now it feels like ARM is trying to pull various coins out of various vending machines.
I think Qualcomm already charges for value on its own customers, eh? I recall something about them taking royalties from device makers as a percentage of the full device price, not a fixed cost per chip. e.g. https://www.cnet.com/tech/mobile/apple-sues-qualcomm-for-1b-...
They do indeed. but their argument was/is that their CDMA licenses made it so that your phone/tablet/etc design was feasible at all. ARM could make this claim too but in ARM's case, it doesn't have standard-essential patents required for network interoperability. That's the bit that pissed Apple off so much that they ordered their suppliers to stop paying the royalties.
This is such a huge-scale massively-disrupting event.
I hope Intel & AMD try to play. I'd love to see ultra-small x86 cores & cut down Gpus come out to play. I tend to doubt x86 is inherently vastly worse. But it's also hard to imagine them wanting to march into a competitive down market where most phones cost less than the starting prices of their regular cpus.
This is a huge chance for RISC-V but so far it has been an unserious competitor. Nothing big, nothing efficient, nothing on vaguely recent nodes. And there's just so so much more than cores at stake. Power management, coherency fabric, there's so many other things mobile chips are great at, tuned at.
It's just wild that there's literally no one else out there.
Would also like to hear the answer from someone with a better grasp of the subject. I have a crude understanding of how basically a CPU works but the real-deal is probably orders of magnitude more complex and convoluted thanks to optimizations stacked on one another.
Intel's "Horse Creek" RISC-V SoC is build on Intel 4, which is a "vaguely recent node". So recent, it's the first chip made using it. You'll be able to buy a board using those around August or so. Working chips were shown at a conference last September.
Ventana's "Veyron 1" RISC-V SoC uses either TSMC 5nm or 3nm. Those are also "vaguely recent". That's an 8-wide 3.6 GHz core, btw, like Apple M1/M2, well ahead of Intel, AMD, or ARM itself.
As for "Power management, coherency fabric, there's so many other things mobile chips are great at, tuned at", if you are someone who already has such as chip, such as oh I don't know Qualcomm or Samsung or for that matter Rockchip or Allwinner, and you already have ARM-based chips with all that stuff, and you swap out the ARM cores for RISC-V cores -- then you still have all that non-core stuff.
Thusfar there don't seem to be any generally available chips in the world. No one has benchmarked anything modern. We don't know what to expect yet, there are no known quantities, from what I can tell.
I'd forgotten about Horse Creek. It's certainly an interesting future board. I like that it has 10Gbe and some USB3. And PCIe 5.0! That does look modern. Not expecting a high value, a good price, but this could be a good starting place & we should learn!
I kind of though Intel had canceled all their RISC-V stuff, but this SiFive board is likely unaffected. However the rest of their once ambitious RISC-V foundry assistance program seems to have been recently killed off, that's why I (accidentally) ruled them out. https://www.theregister.com/2023/01/30/intel_ris_v_pathfinde...
I think the vast majority of the chips out there use ARMs CoreLink and Cache Connect Interface. There's a ton of platform, tuning & fabric that I still suspect is missing from most chipmaking orgs, that's going to require intense redevelopment & which will take lots of release iterations to become decent.
Ventana is a good mention. They don't seem to be shipping at the moment, don't seem to have anyone posting benchmarks on chips they have. Still to me an unknown quantity, but yeah totally as modern a process as we can expect, very exciting.
Intel killed an IDE that was redundant with other existing IDEs.
Nothing at all to do with Horse Creek or Intel Foundry Services as a chip-making service.
No one is shipping really high performance RISC-V yet because it takes 3 or 4 years from project start to having working silicon and it is less than four years since the initial RV{32,64}GC ISAs were frozen and ratified in July 2019.
Lots more stuff extremely important for mobile to servers to supercomputers was ratified in November 2021, including the Vector ISA, cache control instructions, hypervisor spec, bit manipulation instructions (clz, popcount and things like that), 16 bit FP, crypto in scalar registers (SHA, AES), PAUSE instruction, page-based physical memory attributes (e.g. cacheable or not).
That's less than a year and a half ago.
Cores and chips are in the pipeline, being made by credible people who WILL make what they say they will, but it all takes time.
Meantime, the THead C910 core -- first announced in July 2019 -- which is roughly equivalent to the ARM A73 or A73 was available (as usually happens) on an expensive dev board in late 2021 and is coming out in quad core 2.5 GHz form on a $99 board in April (probably). SOPHGO have recently shown a board with the SG2042 chip with 64 C910 cores running at 2.0 GHz. I expect to have access to one of these boards shortly.
Of course these boards, using cores designed in 2019, use the RISC-V specs that were available in 2019, including an early draft (0.7.1) version of the Vector extension which is very similar to but incompatible with the eventual 1.0 ratified in November 2021. Not completely incompatible -- common byte-oriented functions such as memcpy, memcmp, strlen, strcpy, strcmp etc are binary compatible between 0.7.1 and 1.0, and most other things are easy, trivial, modifications. Much much easier than converting between AVX and NEON or between SVE and RISC-V V.
Expect expensive low production volume chips and boards with RVV 1.0 next year, and cheap mass production ones in 2026.
That's how the chip industry rolls. There is a very long production process. But these things are in the pipeline, they are coming, we know what they will do when they get here, and there is virtually zero chance that they won't eventuate.
I like your use of time-scales to try to explain why we are where we are. I hope very much you're right that what we see now is isn't really reflective. That a ton of really big improvements to RISC-V cores are being lined up, will show up.
I'm not going to judge too quickly. But man, waiting till 2026 feels hard. Whether we have any intermittent experience of RISC-V closing the gap in that time is still in doubt to me. And I keep worrying that we'll forever see a huge split: a bunch of super cheap super low-end RISC-V application-processors that are just not good, and a bunch of many-thousand-dollar fancy units, that may or may not get actually good benchmarks/availability. With ARM potentially getting absurd & stupid, the good value for a sub $100 or sub $300 SBC seems like the real competition, and I just have such concern RISC-V is going to bifuracte between extremely up-market & extremely cheap crap.
This might be tolerated in the short term, but the main story here is that their customers now have an incentive to find an alternative to ARM. Do you really want to risk your profit margins being held ransom?
Many moons ago, while I was ARM (note the casing), during a presentation given by someone in the upper management said that 'the guy who sells your phone screen guard makes more money than ARM do off a chip in that phone'. The message was that the margins were tight, and it was probably true.
ARM is Arm now. More cool, more public, and a larger employer than it ever was. I hear that they did some pretty big salary correction recently--ARM engineers used to be paid non-competitive salaries, and there were mass exodus. Obviously, they now need to cover this cost somehow.
118 comments
[ 3.4 ms ] story [ 193 ms ] threadThe interesting part (to me) is speculating which parts of the market ARM will choose to focus on. Microcontrollers seem like a lost cause at their price point, but making a shift to higher-margin licensing might be part of a strategy to acknowledge that loss.
Apple's price-point and margins are high enough but they're pretty protective of their margins afaik. I wonder if it's worth transitioning away though.
I think this will definitely boost Risc-V investment and maybe other alternatives, if for no other reason than to have a viable threat to leave ARM if they decide to get more aggressive in pricing.
https://en.wikipedia.org/wiki/Star_Trek_project
https://www.macrumors.com/2012/06/10/a-bit-of-history-behind... the Mac OS X on Intel Project "Marklar"
Possibly but as it’s believed they were lead partner for the design of the ARM64 ISA and seem to be able to add their own extensions (as well as designing their own micro architecture) they already have a lot of control in a way they didn’t with 68k and PowerPC.
Also do you think the they still maintain macOS ports for PowerPC?
We might look for clues in any Apple contributions to LLVM and Darwin for RISC-V or PowerPC.
I went searching if Darwin still supports PowerPC. I found a Reddit thread about compiling Rust code for Darwin PowerPC, but they only talk about old versions of Darwin:
https://www.reddit.com/r/rust/comments/qbok3d/how_difficult_...
It shouldn’t take that long to port darwin to a new arch anyway. And when/IF RISC-V actually becomes even close to being competitive to ARM/x86 it will probably be very different than it is now.
Is there even a RISC-V that can compete against the ARM Cortex-M33? Or ARM Cortex-M0+ ?
ARM has the microcontroller market in a deathgrip. I'm honestly not seeing any competition here from anybody. STM32, NXP, Microchip SAM, etc. etc. Its all ARM.
You get some 8-bitters (8051 lol still alive) at the lowest end with SiLabs USB stuff and other such stuff. But ARM is incredibly dominant here.
The competition isn't here yet, but the bloody strangler is knocking at the door and someone upstairs just yelled "Nobody's home!"
https://www.cnx-software.com/2022/10/22/10-cents-ch32v003-ri...
Robotics and IoT edge is another. The bl808 is a 5 core chip for $2odd with the lowest power mcu having access to 64mb ram, no other mcu can do that as far as I know:
https://m.aliexpress.com/item/1005004970779483.html
There is also more of a drive to open source drivers etc with many of the riscv mcus.
There's probably more parts with data sheets in Chinese only.
Renesas has started replacing ARM core by RISC V in their new micro-controllers. They still offer their ARM products and will do for a long time, but move new ones to RISC V.
ARM's advantage is at the high end. For a high performance core it's key to work closely with TSMC and Samsung on the most advanced nodes, to tune the design to what leading edge processes can do. This is very labor and capital intensive. It definitely gives ARM an advantage against smaller IP providers at the high end. But at the low and mid end? Nope. It's where ARM is easier to replace.
None of these chips you're talking about are on Digikey or Mouser though. I dunno, are they not selling these chips to the big retailers in order to try to cut costs down or something?
Except Renesas, none of the companies mentioned make chips. Neither does ARM.
They license CPU cores to companies that make chips.
If you say there are not many publicly available products including publicly accessible RISC V cores from those companies I listed compared to ARM, yes this is true. Because they're (for now) mostly used for embedded application, deeply inside complex SoC. Only used by the chip makers developers.
A relatively small company like Andes has their IP in over 10 billions SoC today. A lot of it for things like touch screen controllers, but also more complex things now. But I'm not sure you can find a mass product with an accessible RISC V Andes core.
Still, public access has already happened with Renesas, and it will likely get more common. People tend to focus on the high end, but the most common in high tech is to attack from below. Take a foothold in the low end because you're cheaper (smaller companies with lower costs can deal with lower prices), build a customer base, then grow from this. Until one day you can attack the high end (we're not there yet IMHO). I have access to these companies roadmap and this is what I see them doing, and it shouldn't be surprising.
But this means that at this stage, the majority of RISC V use is invisible to the public. To me this is perfectly normal. Building an ecosystem and maturing a technology like this takes a long time.
IMHO we're already at a stage where ARM has real competition in the low and mid end, particularly where there's little dependency to the ISA (embedded, or simple uC programmed in C anyway with no 3rd party binaries included, like Renesas). I think ARM is still safe at the high end though, and it will take some time for RISC V to get there. Chip making is hard, particularly at the high end.
Yes, of course, lots of them. That is where RISC-V has taken the most market share from ARM so far.
SiFive's E20, E21, E24 hit various points of that. Plus they offer the S21 which is a 64 bit microcontroller in Cortex-M33 class. ARM doesn't offer anything like that -- if you want 64 bits from ARM then you have to take the full no-subsets ARMv8-A ISA.
The Europeans have their Ibex (formerly Zero-RISCy). Andes has the N22.
I don't think there's even a power benchmark on any of those chips you've mentioned. At least, I'm not seeing any online in my searches.
STM32U5 (and most other Cortex-M33 based microcontrollers) are ~110nA of power consumption sleep mode, and 16uA / MHz at 160MHz, some of the most power-competitive chips I'm aware of on the market. I'm talking "run a microcontroller on a CR2032 cell for 10 years" kinda thing here.
--------
EDIT: Looking at the ULPMark website (https://www.eembc.org/ulpmark/ulp-cp/scores.php), it looks like the #1 chip is RSL15, which is ARM Cortex-M33, as I expected. With a ULPMark score of 1090 (higher the score, the less power the device uses). Nearly everything on the top list is Cortex-M33...
Except LP5100 and XRM32UL051 are some Chinese Chips I don't understand. They got some impressive scores but I don't know what architecture they have. Nearly everything else is Cortex-M33 (except for the Ambiq, which is ARM Cortex-M4).
Cortex-M0+ was the king of this a few years ago, but M33 has displaced it. We still see Cortex-M0+ chips due to much lower costs around though. So if RISC-V is around Cortex-M0+ level, I'd still be interested.
SiFive, like ARM, doesn't make chips, they license cores to people who make chips. The power characteristics are up to the chip designer -- what node they use, which "corner" of that node they use, and various other chip design features -- and not the core designer.
I don't know of anyone who is putting out stand-alone microcontroller chips using SiFive cores. They've been concentrating on getting them into larger systems. For example Qualcomm said a few months ago they've shipped 650 million SoCs (i.e. Snapdragon XXX) with RISC-V cores in them (probably SiFive as they're a SiFive investor).
The closest I know of (using SiFive cores) would be the Bouffalo Lab chips, but they are not just microcontrollers but WIFI/BlueTooth which of course by their very nature use considerable power. The BL602 datasheet says it uses 500 nA in "hibernate" (RTC or GPIO wakeup) and the CPU 22 mA at 192 MHz with radios off, so 115 µA/MHz. Certainly not in the league you quote.
Well, actually ... I just looked and the STM32U5 datasheet says 440 nA "standby" with RTC or GPIO wakeup, so actually 440 vs 500 is not all that different, comparing apples to apples.
No they don't actually. RISC-V is already selling billions of devices and its growing at an absurd rate.
You have companies like WesternDigital transforming all their products to RISC-V. And a lot of harddrive and SSD vendors are following. You have companies like Andes that push RISC-V onto every cheap consumer device from China. Companies like SiFive are pushing into the automotive market. NXP has RISC-V products already. Companies like Gaisler into the space market.
RISC-V is in more places then people realize and its still very early in the adoption curve. There are like 100s RISC-V companies pushing RISC-V into every niche.
> RISC-V that can compete against the ARM Cortex-M33? Or ARM Cortex-M0+
In terms of performance, RISC-V offerings are already beating them. ARM has momentum but in terms of performance per area RISC-V is better.
Can you please give me a link to Digikey (or Mouser) with the part number that I can use to verify this claim?
Here's what shows up in my search: just PowerPC and ARM chips from NXP: https://www.digikey.com/en/products/filter/embedded/microcon...
--------
Note that Cortex-M33 parts are extremely competitive. See my discussion points here: https://news.ycombinator.com/item?id=35296747
As for the Cortex-M33, I don't know if RISC-V cores in your link. So not sure we have a great bases for comparison.
If if it is better now, there are so many more companies competing on core design, I don't think ARM can consistently outperform all of them in all segments. RISC-V is just a superior business model and it will eventually just win.
In my experience, bet on legacy. There will always be enough time to wait and switch later if things turn out better.
This is doubly so in embedded, where legacy issues / manufacturing issues reign king over performance.
-------
Cores are often the wrong thing to focus on anyway at the embedded level. I don't think performance is a major concern. Maybe power/performance is a concern in battery constrained apps, but many electronics will bottleneck elsewhere.
So I'm really not sure how or why RISC-V would break into the uC world / low end.
If anywhere, they have a chance at the higher level, like at Rasp Pi level chips or stronger.
Performance isn't that important maybe (depends on the application), but energy often is, size is, having lots of vendors is, having a large software base is.
> So I'm really not sure how or why RISC-V would break into the uC world / low end.
If by lower end you mean 8-bit, then no it wont. But one of the large adopters of RISC-V are actually places where there used to be tiny 8-bit cores and now people want more substantial 32-bit cores and RISC-V is really good there.
And it has already broken into that market as there are already lots of RISC-V products shipping, lots of them in lots of consumer electronics from China.
RISC-V will continue to grow in that market, if you were a MIPS costumer for example, you are likely gone move on to RISC-V?
> If anywhere, they have a chance at the higher level, like at Rasp Pi level chips or stronger.
At the very low end being very cheap, not involving as many lawyers actually matters quite a bit because volumes are high. Specially when there are very competitive verified opensource cores as well.
Ibex: https://github.com/lowRISC/ibex
Used by Google in the OpenTitan Project
Open Hardware Group:
https://www.openhwgroup.org/
Chips Alliance:
https://www.chipsalliance.org
Not to mention all the commercial offerings (or commercial offerings that support those open-cores). They might not be open, but its easy to get up and running with RISC-V compared to commercial stuff.
There is just so much education moving to RISC-V. Lots of people will be educated in RISC-V and lots of people will look at some of these OpenSource cores to learn. When places like Berkley, Stanford and ETH embrace something it usually trickles down to most universities.
Outside of application cores though, no real moat.
The software isn't ready and they are primarily not looking at smartphones yet.
It will take a few more years until we get there, but once the software is ready you will have lots of companies competing.
CPUs are amazingly complicated. A major new CPU designs can take 3-5 year, consume many $100M, and end up not competitive. Even Intel has had serious issues after many years dominating their market.
The ecosystem of OS kernels, platforms, testing/verification tools, compilers, debuggers, and related can take awhile to mature. Android is a major platform for phones, and while by default apps are portable, assuming the vm is ported, but some apps write native code to avoid the Java related overhead. Intel tried a phone and ended up emulating ARM chips for native arm code and delivered terrible performance and terrible perf/watt. That project died with a total design win of (I believe) just a single relatively low volume phone.
So making a CPU to compete with a flagship android phone isn't easy, and even if investing some multiple of $100M and years you might end up still behind, or bankrupt when you scare Arm into cutting prices again.
Device maker: I'd like to buy some ARMs please.
Arm: for what final product?
Device maker: none of your business.
Arm: we're making it our business. Tell or get lost.
Device maker: if you must know, we put them in our Mars rocket...
Company: how much money does your product cost?
Seller: how much money do you have?
Hmm, I wonder if this could put the brakes on the vomitously expensive phones that dominate the market these days.
Soon everyone from CPUs, GPUs, memory, etc will be trying to charge a recurring fee.
It's good news for RV, even more reasons to ditch ARM
I suspect we are now in a (possibly short) period when there is no viable alternative to Arm on mobile say.
When that period ends then the alternatives (RISC-V based) won’t be free. There will still be a cost.
This move obviously looks like a move to juice profits short term ahead of a float, but perhaps also to tilt pricing to keep Arm competitive on the lowest end devices.
What about POWER?
There's this RISC-V dream about an high performance CPU with a fleshed out, open source ISA, with a company championing it with a big proven core you can buy, and decent compiler support in place... but we already have it! Its right over there!
Also, I think POWER is a bad omen. If IBM can't sustain POWER's hype (even with good CPUs you can buy right now), I'm not sure how RISC-V is supposed to surpass that from square one.
I don't want to hate on RISC-V, but I just don't see a line from here to where ARM is now, even with ARM's shenanigans.
Other countries (like Russia, and China's MIPs thing) have pursued novel architectures for higher power general purpose computing, but its mostly for government use I think. China pulling off a societal transition to RISC-V would be quite a coup.
I think the main problem of POWER today is that you can't get a "Raspi sized" development board, while you can easily get all kinds of ARM based SBCs as well as various RISC-V based SBCs. The smallest POWER system you can get right now is iirc a workstation from Raptor which essentially competes with Xeon workstations, if you don't want to play around with soft cores on FPGAs.
https://github.com/antonblanchard/microwatt
- https://github.com/OpenPOWERFoundation/a2i
- https://github.com/OpenPOWERFoundation/a2o
- https://github.com/antonblanchard/microwatt
I don't understand why. RISC-V is being used and pushed by 1000s of companies at this point. You can license cores from 100+ vendors or you can make your own.
There are lots of companies pushing it into every single part of the market. There are well kown cores for space travel, the next generation is RISC-V. RISC-V is already in plenty of low level cores, in things like SSDs, NICs and so on. Companies like Andes push it into billions of consumer devices.
In terms of the ISA itself it is already a match or better then ARM in most ways. And in 1-2 years with the next rounds of specs, it will be more advanced.
To me it seems like RISC-V is already winning by far and its just a question of time. This is more like Linux in the late 90s vs commercial Unix.
But there is basically no reason to pick POWER over RISC-V at this stage. Not unless you are already invested in the POWER architecture.
RISC-V has far more OpenSource work, far more companies involved, the spec is driven and improved by a lot of people and so on.
The reality is most work on RISC-V doesn't go into super high performance stuff and if its there, then usually along side high performance AI stuff. SiFive, Esperanto, Tenstorrent and so on.
> I'm not sure how RISC-V is supposed to surpass that from square one.
Buy having 100x more companies working on it.
ARM is an incredibly important foundational company but they barely make any money, hell they lose money about half the time, including during the pandemic, despite absolutely crazy electronics sales!
Softbank isn't a charity, they want $40b+ valuation on a company that makes $0 +/- 1.5b ish. They want almost double what they paid for it, in fact.
The only companies willing to pay that much money are the ones who see some kind of strategic synergy with owning ARM. NVIDIA could have gotten CUDA into every smartphone's default graphics IP, and become kingmaker licensing one of the most important ISAs in the world. Not only would they not kill that "just to sell some more tegras", but actually it likely would have been a watershed moment of NVIDIA moving towards licensing their IP to external companies. This would have been the start of the pivot towards semicustom - just like AMD already does.
People just don't want to accept that because of "green man bad syndrome". But NVIDIA isn't exactly what people always present them as either - they are moving towards open linux drivers, they opened up to adaptive sync, both of which people would have (and did) defend to the death would absolutely never happen, hell would freeze over. And now that they are, still, nothing is ever good enough - well they have a closed userland, they have blobs (which so does AMD, in linux-firmware), they have a security processor! Even if those things are also true of other brands... the green man bad field means it'll never be good enough.
There were always two alternatives to NVIDIA buying it - another sugar-daddy with business synergy, someone like TSMC ("bundle deal if you get it fabbed here") or Samsung or someone like that that could pair it with some other aspect of their business. Or finally, Softbank just makes it actually turn a profit. Who pays for that? Well, big tech, and probably ultimately consumers too.
Companies like Google and Amazon and Tesla have grown fat on the "pay almost nothing for this ISA, bolt on our accelerator, sell it for billions" while ARM gets nothing and that's never been sustainable in the long term.
Companies like HDMI already do make private tech that is licensed per-device, and again, you can't even have HDMI 2.1 in linux on AMD because it's closed-source and they won't license it in a compatible way, while NVIDIA can offer support because of those blobs. Qualcomm requires similar per-device licensing on their modems too, and wireless/modem blobs are some of the most notoriously opaque/blobby peripherals too. People deal with it, pay the licenses, and move on.
https://gitlab.freedesktop.org/drm/amd/-/issues/1417
And the "big tech forms a consortium and buys it and runs it as a nonprofit" was always a pipedream. Never going to happen, and none of the companies involved have any incentive to push tech forward in the same way as, say, Apple does. They just need it to be Good Enough And Cheap, and it already is. And scraping up $40b to keep it cheap is a bit of a mixed bag.
Well, you didn't get NVIDIA. You did get big ARM layoffs (just as Jensen warned, and just as people scoffed at, ARM UK dropped 20% of their staff) and big price increases. Just as predicted. /shrug
https://www.tomshardware.com/news/arm-uk-jobs-cut
And yeah RISC-V exists and it's good to have an alternative to x86 and ARM too. There's a bit of a lifecycle here where they tend to get stagnant over time, maybe RISC-V being open-source will help that. But it also is going to be hell for ISA compatibility across vendors (a lot of the codes ar...
And we know exactly who Nvidia would have made king amongst all the Arm licensees.
I'd add to your analysis that, given the immediate IPO-driven need to demonstrate they can increase margins, raising prices sooner is probably better than later because RISC-V is not yet a viable competitive alternative for ARM's most attractive customers. However, later it very likely will be and if RISC-V proves to be a threat to ARM's blended margins, ARM can drop prices then because at least the IPO will be over.
I did like 'Arm has been "frustrated by customers’ reluctance to accept the new arrangement."' why ever would that be? why would customers not love this value proposition?
The problem here is that ARM's current owners - SoftBank - are starving and trying to eat their seed corn. The problem with a company like ARM is that's value is mainly that they don't take all the value of the widget. This is in contrast to, say, Intel or Nvidia who completely squeezed all the margin out of the PC business decades ago. Problem is, "we leave money on the table for our customers" is not something the stock market will buy, so they have to start squeezing.
(Same reason why Nvidia tried to buy them.)
Personally I don't see this being fixed unless ARM is either privately owned by someone who just has the money to burn on an FRAND-licensable architecture, or they get nationalized.
https://youtube.com/watch?v=FO5PsAY5aaI
The collaboration with Apple failed spectacularly. The only reason ARM is alive today is because Nokia and Ericsson needed a better CPU for their phones and ARM happened to be really power efficient (mainly by accident, but that's a story for another day).
The 'happened' on power consumption is true on day 1 in 1985, but the Arm team continued to keep power consumption low and added features, such as Thumb, that won Nokia and others over.
Shameless self plug for more on this stage of the Arm story:
https://thechipletter.substack.com/p/the-arm-story-part-3-cr...
It doesn’t _exist_ because of this, but it’s definitely getting a lot of attention because of this.
Of course, there are vendors who are only RISC-V adjacent, but they will either put up, shut up or subsume the entire ecosystem if they get big enough. Time will tell.
RISC-V despite its young age is even more fragmented. This is because the Berkeley folks took their sweet time to finish important parts of the spec and in in the mean time multiple companies shiped out RISCV with home grown ISA changes.
Fragmentation is a feature of RISC-V, and often it represents actual diversity in use, more then just needless fragmentation. And the majority of the time its additions, not changes to anything that in the spec. And RISC-V specifies how that it is be done in order to stay compatible.
In terms of software for Linux its not actually fragmented. Pretty much all major linux and free software distribution agree on targeting the RV-A profiles.
Please don't confuse this with openness
> How many CPUs with a home grown memory protection designs were sold between those two dates?
Zero that I'm aware of. Want to inform us?
The Canaan Kendryte K210 dual core 400 MHz 64 bit microcontroller shipped with an incompatible MMU, but it wasn't home grown — it's unmodified Berkeley Rocket implementing the current version (Priv 1.9.1) as at the time Canaan took the core snapshot for their SoC.
If that chip was a serious Linux target then priv 1.9.1 support could be put BACK in the Linux kernel (it wasn't upstreamed but it existed and was published). But with only 8 MB of non-expandable on chip SRAM there is no point anyway.
Over 10 billion shipped so far. Chances are you're using a few without realising it. Got an Android phone made since 2019? Qualcomm has shipped 650 million Snapdragon SoCs with RISC-V in them. Got a WD hard drive? Got an NVidia GPU made in the last five years or so?
> where can i buy one at a reasonable price?
What are you after, exactly?
50 chips for $5?
https://www.aliexpress.us/item/1005005036714708.html
A quad core 1.5 GHz Linux SBC at near Raspberry Pi price?
https://www.amazon.com/youyeetoo-StarFive-VisionFive2-Comput...
Those are the extremes of the mainstream range you can buy right now. Considerably higher performance is rolling out all year, including many-core (e.g. 64) OoO server chips.
I hope Intel & AMD try to play. I'd love to see ultra-small x86 cores & cut down Gpus come out to play. I tend to doubt x86 is inherently vastly worse. But it's also hard to imagine them wanting to march into a competitive down market where most phones cost less than the starting prices of their regular cpus.
This is a huge chance for RISC-V but so far it has been an unserious competitor. Nothing big, nothing efficient, nothing on vaguely recent nodes. And there's just so so much more than cores at stake. Power management, coherency fabric, there's so many other things mobile chips are great at, tuned at.
It's just wild that there's literally no one else out there.
AMD has been very successful but wouldn’t have the bandwidth to try this.
Is it actually possible to have performant x86 cores for handheld devices, or is that just the limitation of x86 with almost 50 years of legacy?
[1] https://shakti.org.in
Intel's "Horse Creek" RISC-V SoC is build on Intel 4, which is a "vaguely recent node". So recent, it's the first chip made using it. You'll be able to buy a board using those around August or so. Working chips were shown at a conference last September.
Ventana's "Veyron 1" RISC-V SoC uses either TSMC 5nm or 3nm. Those are also "vaguely recent". That's an 8-wide 3.6 GHz core, btw, like Apple M1/M2, well ahead of Intel, AMD, or ARM itself.
As for "Power management, coherency fabric, there's so many other things mobile chips are great at, tuned at", if you are someone who already has such as chip, such as oh I don't know Qualcomm or Samsung or for that matter Rockchip or Allwinner, and you already have ARM-based chips with all that stuff, and you swap out the ARM cores for RISC-V cores -- then you still have all that non-core stuff.
Thusfar there don't seem to be any generally available chips in the world. No one has benchmarked anything modern. We don't know what to expect yet, there are no known quantities, from what I can tell.
I'd forgotten about Horse Creek. It's certainly an interesting future board. I like that it has 10Gbe and some USB3. And PCIe 5.0! That does look modern. Not expecting a high value, a good price, but this could be a good starting place & we should learn!
I kind of though Intel had canceled all their RISC-V stuff, but this SiFive board is likely unaffected. However the rest of their once ambitious RISC-V foundry assistance program seems to have been recently killed off, that's why I (accidentally) ruled them out. https://www.theregister.com/2023/01/30/intel_ris_v_pathfinde...
I think the vast majority of the chips out there use ARMs CoreLink and Cache Connect Interface. There's a ton of platform, tuning & fabric that I still suspect is missing from most chipmaking orgs, that's going to require intense redevelopment & which will take lots of release iterations to become decent.
Ventana is a good mention. They don't seem to be shipping at the moment, don't seem to have anyone posting benchmarks on chips they have. Still to me an unknown quantity, but yeah totally as modern a process as we can expect, very exciting.
Nothing at all to do with Horse Creek or Intel Foundry Services as a chip-making service.
No one is shipping really high performance RISC-V yet because it takes 3 or 4 years from project start to having working silicon and it is less than four years since the initial RV{32,64}GC ISAs were frozen and ratified in July 2019.
Lots more stuff extremely important for mobile to servers to supercomputers was ratified in November 2021, including the Vector ISA, cache control instructions, hypervisor spec, bit manipulation instructions (clz, popcount and things like that), 16 bit FP, crypto in scalar registers (SHA, AES), PAUSE instruction, page-based physical memory attributes (e.g. cacheable or not).
That's less than a year and a half ago.
Cores and chips are in the pipeline, being made by credible people who WILL make what they say they will, but it all takes time.
Meantime, the THead C910 core -- first announced in July 2019 -- which is roughly equivalent to the ARM A73 or A73 was available (as usually happens) on an expensive dev board in late 2021 and is coming out in quad core 2.5 GHz form on a $99 board in April (probably). SOPHGO have recently shown a board with the SG2042 chip with 64 C910 cores running at 2.0 GHz. I expect to have access to one of these boards shortly.
Of course these boards, using cores designed in 2019, use the RISC-V specs that were available in 2019, including an early draft (0.7.1) version of the Vector extension which is very similar to but incompatible with the eventual 1.0 ratified in November 2021. Not completely incompatible -- common byte-oriented functions such as memcpy, memcmp, strlen, strcpy, strcmp etc are binary compatible between 0.7.1 and 1.0, and most other things are easy, trivial, modifications. Much much easier than converting between AVX and NEON or between SVE and RISC-V V.
Expect expensive low production volume chips and boards with RVV 1.0 next year, and cheap mass production ones in 2026.
That's how the chip industry rolls. There is a very long production process. But these things are in the pipeline, they are coming, we know what they will do when they get here, and there is virtually zero chance that they won't eventuate.
I'm not going to judge too quickly. But man, waiting till 2026 feels hard. Whether we have any intermittent experience of RISC-V closing the gap in that time is still in doubt to me. And I keep worrying that we'll forever see a huge split: a bunch of super cheap super low-end RISC-V application-processors that are just not good, and a bunch of many-thousand-dollar fancy units, that may or may not get actually good benchmarks/availability. With ARM potentially getting absurd & stupid, the good value for a sub $100 or sub $300 SBC seems like the real competition, and I just have such concern RISC-V is going to bifuracte between extremely up-market & extremely cheap crap.
Android supports x86_64. Intel and AMD do play in the mobile space. But their smartphone-oriented designs just couldn't cut it.
They absolutely wanted to be a part of the largest market of CPUs in the world.
ARM is Arm now. More cool, more public, and a larger employer than it ever was. I hear that they did some pretty big salary correction recently--ARM engineers used to be paid non-competitive salaries, and there were mass exodus. Obviously, they now need to cover this cost somehow.