I’d also add that SemiAnalysis posts a lot of conjecture as fact and loves sensationalism. In this case it’s taking Qualcomms word for it and not representing Arms own dispute of the claim.
So the original source for all of these articles is flawed to begin with as well. I would describe SemiAnalysis as a technical tabloid. Like The Daily Mail, there’s some small truth to the claims but the rest is just made up for clicks.
The title should really be 'Qualcomm claims ARM to prohibit proximity of CPU w 3rd-party modules in one chip'.
As others have pointed out Arm disputes these claims.
As to the wider dispute Nuvia developed technology using Arm IP under one license and then Qualcomm want to sell that technology under a different - and presumably more favourable and lower cost - license.
The big risk for Arm was always that a big architecture licensee becomes dominant and damages sales of Arm's own cores. That would not only be bad for Arm but bad for customers who use Arm's cores. Whatever you think of the tactics I think this is what Arm is trying to stop Qualcomm doing here.
I interviewed at one of Qualcomm's satellite locations, and when you walk into the receptionist's desk area, the walls are totally covered in golden plaques. Floor to ceiling, wall to wall. Each golden plaque is one of their patents.
This is sadly not all that unusual at US technology businesses, you will see it at far smaller firms too.
Even Tesla had a patent wall in their lobby until their famous "we don't do patents" post. I've seen the same thing at multiple shitty enterprise software firms as well. Some businesses even demonstrate their spurious number of patents issued annually as some kind of metric for innovation - always a bad sign.
I've also been on the receiving end of the annual "we want to pump the patent count metric" email, when company wide email just blasts for suggestions of anything whatsoever we could patent to up a pointless integer.
Yes, of course. But humans are territorial and possessive, and it aggravates us when someone “steals” our ideas. The iterative and derivative nature of those ideas notwithstanding.
I think ARM is upset their NVidia buy-out was cancelled. This essentially forces someone else to purchase the company, as the total license fees for a company like Google/Apple could outstrip the straight acquisition cost. It's the only plausible explanation I have, as this is basically a giant foot-gun set on destroying the ARM ecosystem right as it's becoming ubiquitous.
Either that or the MBA's have taken over, the engineers have left the building, and there's no innovation left in their product roadmap. If there's no future in their IP then ARM is just a glorified patent troll; in which case I'll gladly watch them burn down as RISC-V takes over.
> This essentially forces someone else to purchase the company, as the total license fees for a company like Google/Apple could outstrip the straight acquisition cost.
The value of fees paid by these companies is nowhere near the value of Arm.
> If there's no future in their IP then ARM is just a glorified patent troll
Seriously? Qualcomm wants to use IP that was 100% developed by Arm and is used in billions of smartphones and other devices around the world. You may not think that it has value but Qualcomm clearly does.
>IP that was 100% developed by Arm and is used in billions of smartphones and other devices around the world.
You mean the ISA? Qualcomm has a license to the ISA that allows them to make and sell custom microarchitectures. They had it before they acquired Nuvia.
>You may not think that it has value but Qualcomm clearly does.
Qualcomm will likely push RISC-V hard, thus not use ARM ISA at all, from now on.
Refer to this recent announcement by SiFive[0] for some Qualcomm quotes that hint at that.
>"We are excited to see RISC-V solutions for wearable and consumer devices becoming a reality, and we are looking at possibilities of integrating SiFive’s latest products into Snapdragon platforms,” said Ziad Asghar, Vice President, Product Management- Snapdragon Technologies and Roadmap at Qualcomm.
edit: I meant "Qualcomm will likely push RISC-V hard, thus not choose ARM ISA at all, from now on." -- My bad.
i.e. they will not be using ARM by choice anymore.
Unfortunately, I cannot edit the parent post anymore; As it's often the case, it's easy to miss things even when re-reading your own comments.
It was blatantly obvious when reading it again now.
To get back on topic... you said:
>Qualcomm wants to use IP that was 100% developed by Arm
And then I told you,
>You mean the ISA? Qualcomm has a license to the ISA that allows them to make and sell custom microarchitectures. They had it before they acquired Nuvia.
Is there a good overview somewhere of what the realistic path to market is for it? I don't mean this as sass— I'm genuinely curious. As far as I'm aware there are only a handful of real chips/devkits available at this point; mostly it's still people synthesizing cores on FPGAs. Basically the whole story looks pretty experimental.
I know "can I buy it on Digi-Key" is irrelevant to an entity like Google or Apple, but I feel like hobbyist accessibility is at least something of a proxy for general readiness.
There are a quite a few RISC-V chips consumers can buy right now, but mostly MCUs (which are the majority of the market, just not the highly visible or exciting part of it). Higher-powered "PC-grade" chips and dev-boards are coming slowly, and the next year or so should bring a few new entrants, but I wouldn't expect to see a lot of them. Server/mainframe class probably won't be seen for a long time, if ever. It obviously takes longer to develop bigger chips, as well as being a far less attractive market for new entrants.
This last article says, "Since the storage company isn’t a traditional IP vendor, it has worked with Codasip to support its CPU licensees. Including [WD's] Swerv cores and other open-source designs, Codasip has helped customers ship two billion RISC-V chips over the past few years." (Emphasis mine.)
But I think instead of "an entity like Google or Apple", or even WD, you should be thinking "an entity like Baidu, Huawei, or AliBaba". The strongest RISC-V story is a Chinese national security thing, and it works like this.
— ⁂ —
Most computer hardware has been designed and assembled in the PRC for several years now. In 02019 the USG imposed heavy sanctions on a number of PRC hardware companies, including https://en.wikipedia.org/wiki/Huawei, the largest telecommunications equipment manufacturer in the world and since 02020 the largest smartphone manufacturer in the world. These sanctions deprived Huawei of access to chips from TSMC in Taiwan, even though Taiwan is nominally not under US jurisdiction. Computer hardware relies heavily on trade with Taiwan and other countries, including, in particular, MCUs and CPUs fabricated at TSMC in Taiwan using designs licensed from ARM. ARM China, a former ARM subsidiary, has the legal right to license ARM designs for use within China but not overseas.
Using a component makes you vulnerable to backdoors inserted by the component's manufacturer or, if you do not understand its design, its designer. Such backdoors are known to have been a major instrument of US foreign policy in the past (the Crypto AG affair) and presumably still are. The Huawei sanctions were justified by allegations that the PRC is doing the same thing.
> [the WCH CH32V003 is] quite timely as we were forced by HQ last Friday to revise all product line designs to comply with 100% Chinese domestic semiconductors & passive components by latest Q4 end. All western sourced/accounts and parts are now forbidden, apart from those sourced by HQ’s registered jurisdiction.
> Worth adding to the list, we’ve also signed accounts with GigaDevice and Expressif Shanghai in the MCU and connectivity range.
— ⁂ —
So that's the problem to solve.
RISC-V already offers PRC manufacturers a path to building ARM-competitive MCUs and low-end CPUs which can be legally sold overseas and which comply with this onshore mandate. As a bonus, they can be sold at prices several times cheaper than any ARM alternative; WCH's CH32V003 is reputedly 10¢, though LCSC doesn't seem to have them in stock yet. Presumably, until SMIC (or another domestic alternative) can move up to more modern process nodes, fully-sanctions-proof versions of these chips will be larger, more expensive, and more power-hungry than TSMC-manufactured chips. But for many applications this doesn't matter.
The Allwinner D1, using a RISC-V design from the Honey Badger open-source hardware subsidiary of AliBaba, is already performance-competitive with early Raspberry Pis, though I think that performance relies on TSMC. In 3–5 years RISC-V might be able to move up to today's desktop or laptop CPUs, and SMIC might be able to fabricate things at early Raspberry Pi performance levels (22 nm? Something past the end of Dennard scaling, anyway).
— ⁂ —
Also, though, you shouldn't totally dismiss the FPGA cores. FPGAs are a lot cheaper than they used to be, thanks to Lattice. It's true that an ASIC MCU will always be faster than an FPGA fabbed in the same process running a softcore, but it may not have the peripherals you need, and there are a lot of applications where we run an MCU at a 32768 Hz clock to save power, so clock speed isn't everything. Olof Kindgren's bit-serial RISC-V core SeRV https://github.com/olofk/serv reputedly fits into 200 4-LUTs like those in the Lattice iCE40LP384, which would cost US$1.82–2.15 at Digi-Key if they had any: https://www.digikey.com/en/products/detail/lattice-semicondu... and that would leave you another 184 LUTs for peripherals. You can get MCUs a lot cheaper than US$2 but US$2 is still within the budget for a lot of projects.
Every year there are more and more custom ASICs from more and more companies, and a lot of them include some kind of microcontroller core to orchestrate their activities. Historically this has often been something like an 8051 or Z80, but switching it to RISC-V seems likely to lower development costs significantly, especially if a tiny RV32I core like SeRV or the smaller configurations of PicoRV32 would be adequate.
> In 02019 the USG imposed heavy sanctions on a number of PRC hardware companies
Using long now years really makes this harder to read. When a year is within text, there really isn't any reason to format it to be sorted by a machine. No one is confused by the year 300 when they see today so why would some future person be confused when they see 2019?
> ARM China, a former ARM subsidiary, has the legal right to license ARM designs for use within China but not overseas.
FWIW, Allen Wu was finally ousted by leveraging that right to license, i.e. SoftBank's threatening to not release new IP rights to ARM China. SoftBank was somehow further able to wrest back control of ARM China from Wu by cleverly transferring its stake to a SPV jointly owned by SoftBank and ARM, so ARM China is considered a "financial investment interest" rather than a subsidiary.[1]
Hobbyists have a few very cheap RISC-V things they can buy. Check out boards with the Espressif ESP32-C3 or Bouffalo Lab BL602, for example. These are both user-programmable WIFI/BT chips running at 200 MHz or so and with a couple of hundred KB of RAM. They sell for a couple of dollars.
There are also 1 GHz 512 MB Linux boards similar to Raspberry Pi Zero selling for $15 to $30, depending on features.
According to RISC-V International back in June, 10 billion chips have shipped with RISC-V cores in them. Several individual companies have stood up and said they (or their clients) have shipped in the billions each.
It often takes 4 to 5 years from a decision to make a CPU core, to having a design that works in simulation (and on FPGA), to having a working test chip, to fixing any minor bugs and moving into mass production, to having the chip integrated into someone's product/board, and manufactured and delivered to customers or shops. Especially as you move up the performance and complexity scale. This is not just for RISC-V but for anything.
The first ever RISC-V core, in a chip, on a board went on sale in December 2016, less than six years ago. That was the Arduino-style HiFive1 with a 320 MHz 32 bit E31 core (Berkeley Rocket, basically). It performed competitively with ARM Cortex-M3 devices at the same MHz -- except those topped out at 180 MHz. About 15 months later (about 4 1/2 years ago) the 1.5 GHz single-issue quad core 64 bit 8 GB RAM Linux-running HiFive Unleashed went on sale. Benchmarks (e.g. https://hoult.org/primes.txt) put it within about 10% of a DUAL issue ARM Cortex-A53 (in e.g. Raspberry Pi 3) on a clock for clock basis -- except, again, it could run at higher MHz.
Companies that made a decision to create high performance (e.g. wide OoO) RISC-V cores in early 2018 when the HiFive Unleashed came out would likely be JUST NOW getting close to releasing a product for end users.
Except most such companies would have made the decision several years later than that.
Rivos, for example, was founded by ex-Apple A-series and M1 ex-PA Semi people in May 2021, three years after the HiFive Unleashed came out, and about the same time as the Unleashed's successor, the dual-issue HiFive Unmatched came out.
This month, the first mass-produced low cost quad core RISC-V board, the VisionFive 2, is starting to ship to early customers at near Raspberry Pi prices. It is a quad core dual-issue 1.5 GHz 64 bit Linux-running board with on-chip Imagination Tech GPU, on-board M.2 slot for SSD (or SD or eMMC if you prefer). Standard retail price is $55, $65, or $85 depending on whether you want 2, 4, or 8 GB RAM. Early-bird pricing was about 20% less. Performance is about the same as an ARM A-55 running at the same clock speed, or about 80% as fast as the older OoO A72 cores in the Raspberry Pi 4. However the GPU is up to 4x faster than the GPU in the Pi 4, so the desktop experience should actually be better, at least once the software matures.
In September Intel demonstrated a board with their RISC-V "Horse Creek" chip running at 2.2 GHz. It offers performance in the same ballpark with the ARM Cortex-A76 based RK3588 boards that started selling for $150 to $250 back in April or May. SiFive has said that their (not yet named) successor to the HiFive Unmatched will be using this chip. I would not be surprised if SiFive says more about that tomorrow in their presentation at the Linley Fall Processor Conference. In any case it will be just a few months away.
RISC-V chips with performance similar to the Apple M1 (and Intel 10th gen and AMD Zen 2) are in the pipeline.
ARM is terrified of RISC-V and this move seems to be an attempt to prevent anyone doing a lift and shift away from ARM.
RISC-V is a minimal CPU as a manager for a lot of co-processors hanging off it.
What ARM has said is "you can't put your own custom co-processors near an ARM core if they duplicate functionality you could license from ARM". This would prevent you from designing a CPU with an ARM core and a ton of custom co-processors, and then swapping out the ARM core for a RISC-V.
ARM likely had no intent to apply this 'all-or-nothing' strategy to any other licensee, given none of the other architectural licensees are likely to be acquired. Nuvia was somewhat unique in that regard. And the need for ARM's approval to transfer Nuvia's architectural license gave ARM the opportunity to force concessions from Qualcomm.
Precisely. Also I can see the Nuvia license being structured to suit the Nuvia situation - perhaps low upfront fee and higher ongoing fees plus lots of assistance from Arm. Then Qualcomm tries to get out of paying the higher ongoing fees by selling Nuvia cores under their license.
Behind ARM is an venture capitalist investment firm called SoftBank who's intent on making a huge windfall on the sale of ARM either from a takeover or an IPO.
Now that their plan for a sale to Nvidia has been foiled their looking for other ways to squeeze more money out of the pockets of their customers. Probably to increase the valuation of ARM before they take it public.
The only thing this will result in is an increase in the take-up of RISC-V, speeding ARM's demise.
Or maybe this is simply Softbank's plan: burn more brightly one last time before petering out.
SoftBank is not well regarded. They put a lot of stupid money into not great companies. They are generally viewed like the Saudi sovereign fund - dumb money but big.
VC funds depend on a few massive successes, but the majority of their investments are expected to flame out. You cannot judge a fund by its glaring investment failures!
But 95% of VC funds don’t make enough to beat their risk (see link in <https://news.ycombinator.com/item?id=32773493>). Conversely, even that 95% number only implies that VC funds are poor investments: perhaps a 5% chance to win 100x is worth a bet?
Our intuitions are usually wrong when it comes to power laws.
Apple pays for architectural licenses, but doesn't pay for core designs. Regardless, the ARM ISA is not free even if you design your own processor for it.
Why wouldn't Apple pay license fees? They're no longer a shareholder, even if they were I'm sure other shareholders would be upset (ie litigious) if Apple got special treatment.
Apple may have gotten a good deal based on the fact that they don't intend to sell their ARM designs outside of their own products, thus avoiding competition with ARM, but there is no rational reason why ARM would not try to extract as much money out of them as possible.
Also simply because Apple is a big boy. They've historically been able to throw their weight around, for example manufacturing in China & blacklisting when they don't get deals they want/crazy deadlines and pace requirements; remember the suicide nets?
You are correct. The innovation is gone. It has been all MBAs and lawyers and highly diligent immigrants who ask no questions and only do what they are told. It’s a dead company.
As selfish as this is (if it plays out the way the article suggests), it just further reinforces the pattern we see over and over: every spec, every standard, every intellectual property has a poison pill that prevents us from progressing.
I just keep wondering how we can all come together and pull the plug on this stuff. I want a computer built of at least 1000 (RISC-V, MIPS, I would have said ARM before this announcement) chiplets to process the workloads I've explained in my comments since I joined this site. The future of computing is humans cultivating algorithms rather than writing them directly. That looks like genetic algorithms and simulating as many permutations as possible so we can pick and choose the ones that work. And better testing paradigms and formal verification. No more getting lost in the weeds having to learn yet another 3D rendering library or AI runtime. Although the computer I'm talking about can certainly simulate that stuff.
I got my computer engineering degree in 1999 so I could do something like the above paragraph, only to see every attempt to do so be stifled by the main players every step of the way for the last 20+ years. Nobody gets it. They're never gonna get it. They're gonna proprietarize and cripple and dilute every last magical idea until some Evil Corp somewhere can make a dime on it.
Upon rereading the comments, I realized that it might still allow a chip with an array of the same core like I'm suggesting, which would fit with the architecture I've been proposing. Even if that's the case, I'm still against ARM's position. They're basically trying to control how we use their core in something we own, which is the hardware equivalent of DRM.
Have any other SoC vendors come forward on either side of the matter? I don't put it past a pre-IPO ARM to do something like this to make their numbers look better. But I also don't trust Qualcomm to be entirely truthful.
It sounds more like Qualcomm making preposterous claims to avoid having people looking closely as what’s happening with the Nuvia IP. It seems to be working.
That would be classic Qualcomm. As a rule, when Qualcomm disagree with another company, you can assume they are in the wrong.
This is going to seriously slow adoption of ARM. Right as ARM is becoming truly competitive against x86, ARM appears to have lost their minds. By shrinking the number of implementers, ARM is going to lose serious market share, and while NVIDIA and Apple won’t immediately be affected, they’ll likely start pursuing RISC-V. I know Apple already uses RISC-V in non-consumer facing parts of their machines… but we may see yet another transition within 10 years if ARM don’t change their tune.
The litigation between Arm and Qualcomm has taken an unexpected turn.
The British developer of processor architectures decided to radically change its business model: first, license fees will have to be paid by manufacturers of end devices, including smartphones and tablets; second, third-party components, including GPUs, NPUs and ISPs, will be prohibited in chips with Arm processors.
In late August Arm filed a lawsuit against Qualcomm. Qualcomm absorbed server processor developer NUVIA and, according to Arm, had to update its license agreement because it deemed all previous agreements with the company invalid after it bought it.
Qualcomm filed a countersuit against the British developer, and new documentation on the case contains crucial details.
Qualcomm's updated lawsuit says that after 2024, Arm will stop licensing its processor architecture to semiconductor component manufacturers - payments will be charged to end-device manufacturers.
Arm, according to the U.S. company, has already told OEMs that soon the only way for them to get Arm-based chips will be through direct license fees, and they'll have to accept these rules of the game.
But that's not all.
The British company has also decided to tighten its policy with respect to chip developers like Qualcomm: they will not be able to use third-party components in single-chip platforms with Arm processors if Arm offers their analogs as a licensed product.
This will affect graphics and network processors, as well as image processors.
In other words, Arm will prohibit the creation of duos like Samsung and AMD chips, as well as MediaTek and Imagination - both pairs of companies cooperate in the field of mobile graphics.
And Qualcomm itself does not use GPUs from Arm, but its own. Such an initiative Arm has clear signs of anti-competitive behavior, notes the resource SemiAnalysis, and it is possible that these steps will accelerate the companies' efforts to develop chips based on open architecture RISC-V.
However, some of the British company's partners may not be affected by the new rules.
For example, NVIDIA has a 20-year license for the development of components with Arm-architecture.
Apple was at the origins of Arm, so it is unlikely that this tandem can be destroyed so easily.
There is also a version that a mutually beneficial cooperation binds Arm with Broadcom.
Thus, the dispute between Arm and Qualcomm in just over two years threatens to affect the interests of many smaller players.
My own conspiracy theory: Since Nuvia is led by former Apple engineers and Apple doesn't want Qualcomm to have a toe-to-toe fight, Apple pressured Arm to either keep getting the high premium license fee from Apple and destroy Qualcomm and consequently relationships with all other licensees (which includes but not limited to Qualcomm, Samsung, AMD and even embedded market), or Apple would move away and start their new architecture in secret and ditch ARM architecture in long term just like Apple ended its course with x86.
This is a very good news for RISC-V and with the recent news of Android RISC-V port upstreamed to AOSP [1] I will say 2024 (not until 2023 Q4 I guess) will be the year RISC-V starting to explode.
I hope for that!! But will it be better in regards to not every instantiation needings its very own tiny Linux kernel patch and device trees? Or similar one level down into the embedded world, a true CMSIS like HAL that includes peripherals, not the half-assed attempt of ARM-CMSIS where you still always need to fiddle with vendor's peripheral driver SDKs? That would be so great!
I expect that some RISC-V implementations will always need patches, but as any ISA matures it tends to get standardized and the implementations start to look more and more similar. In the case of RISC-V, the costs for both the implementer and the customer will get better as it becomes more standardized, which will help to push both more implementation and more adoption. This in turn drives more standardization… it’s a feedback loop. Then, you will always have your one-off oddities. These are important. Sometimes, semi-custom platforms are for experimentation, and sometimes they’re just super niche use-cases. RISC-V is great in these situations due to zero license fees.
It is not necessarily zero license fee because you can theoretically spun-off RISC-V designs to your own proprietary ISA. Some Chinese scientists did a case study on this possibility by creating a new ISA called RISC-X [1], but they are not progressing to do any harm right now.
RISCV isn't that standardized that there won't be a plethora and mess of peripherals available from every manufacturer. It will be much like ARM.
In many ways it's a good thing. There's no point in more than one manufacturer existing if everything was standardized down to the literal register map. They all end up using the same fabs anyway.
>RISCV isn't that standardized that there won't be a plethora and mess of peripherals available from every manufacturer. It will be much like ARM.
Hardware, sure. Although not as much, as standarization goes further than ARM, with the platform profiles including e.g. standard serial ports, interrupt controllers and so on.
In practice, on the software side, it'll be closer to your usual IBM PC clone.
The boot process, with SBI and UEFI standarized, is also part of the relevant platform profiles. How to get to the point where a kernel loaded in memory is defined. The environment this kernel finds is also defined. The kernel does actually get a description of the hardware.
ARM did very little of that, and did it too late. RISC-V has it all in place before the hardware is even there, so it should be alright.
Signs so far point to a big fat no (as someone running a custom uboot build on visionfive that works better than the bsp from manufacturer). Plenty of blobs, still device trees, etc.
Apple is not about to change their entire architecture that runs across watches, phones, monitors, computers, pencils
, headphones, speakers, and set top boxes over something this minor.
1. Arm will stop licensing its processor architecture to semiconductor component manufacturers by charging payments to end-device manufacturers.
2. Only way for OEMs to get Arm-based chips will be through direct license fees
3. Third-party components such are custom GPUs and NPUs are no longer directly incorporable with Arm processors. This would probably mean they need to operate as a coprocessor of some kind that would not be tightly integrated.
3.1 This means Samsung's Exynos, AMD's PSP and Imagination's PowerVR are probably *fucked*
Conclusion: ARM is really fed up with Qualcomm and is wishing to step up the game by hurting itself. But that said, it's more like "Friendship ended with Qualcomm, Samsung and AMD (but it's more like a frenemy), now Apple and NVIDIA are my homies"
Your post is missing a big "according to Qualcomm, a company known for stretching the rules and truth to favour itself and currently being sued for its misuse of ARM intellectual property following the Nuvia acquisition" disclaimer at the beginning.
I would need at least some OEMs confirmation and a statement for Samsung before I believe it.
I'd normally agree with you, but there's some other data that ARM has been cracking down on third party IP vendors for years.
Imagination is almost dead not just because of Apple opening a chip dev branch basically across the street and offering Apple salaries to nearly anyone who wanted to jump ship, but also because of ARM licensing Mali + ARM CPUs for cheaper than ARM CPUs by themselves. It's why you see Mali in just about every Mediatek and Allwinner design. Even a free IMG GPU would be more expensive. It's why IMG went through that boondoggle buying MIPS in the first place.
So Qualcomm's complaint matches the current trajectory of ARM's corporate strategy.
It's possible that Samsung got a briefing on what's happening with ARM and Qualcomm and negotiated an extended reprieve from the new licensing scheme about a month ago when Masayoshi Son had a high profile meeting with the head of Samsung to discuss "a strategic alliance between Samsung and Arm."
Eh. The ISA of RISC-V is free and open. Actual implementations of RISC-V are not, in general. The point of RISC-V is that you may have the option to switch implementations if you become fed up with your current supplier, without having to switch ISA.
First of all: the licensing of RISC-V is much cheaper than ARM. I've seen Chinese RISC-V MCU's that only cost a dime or less. And these aren't itty-bitty 4-bit parts mind you, but comparable to ARM M0 or M4.
Secondly, there are numerous open-source RISC-V designs out there to give any company a head start and the freedom to modify it as they see fit.
Those OSS RISCV designs are academic largely. True performance RISCV designs are going to be people's money maker and never open sourced.
A head start is great but that still costs expensive skilled engineering time to actually make a worthwhile product. And due to licensing of OSS, it may be a bad thing for a company to build their multi-billion dollar business on something they need to open source for someone to immediately knock them off and kill their business.
In the hardware world, we live and die by our design secrets. If someone can simply copy you exactly, they can completely undercut you and basically put you out of business. A bit different than the software world.
> And due to licensing of OSS, it may be a bad thing for a company to build their multi-billion dollar business on something they need to open source for someone to immediately knock them off and kill their business.
Nobody, other than a company that makes only CPUs or CPU IP, adds value in the CPU. If you're anybody else, a CPU is a commodity dingus that you need to run your software, which is where the interesting things for your billion dollar business go on.
If you are a large user of CPUs, or even CPU IP, there's only so much you're going to pay before you band together with other large users and cut out the person collecting the payments. The freedom to make small tweaks may also be a a nice extra for a very few users. Maybe you need a special functional unit to speed something up. But your core value is not going to be in those tweaks, and in fact you save money if you can upstream them and get somebody else to do some or all of the maintenance.
That's what happened with the operating system layer. Nobody worries that they're "building their business" on the Linux kernel.
The CPU/GPU/Memory SoC in a Mac is a line item, albeit a hefty line. The one in an iPhone is more so. They sell systems as their product, not retail box processors. It's thought an M1 cost them around $50 in a $2000 laptop.
Well, I think we're of two valid points of view on this one. You're saying Apple adds value to the CPU itself, which taken abstractly is true since they do have some advancements in the CPU and how they package it in the SoC. I'm saying that concretely they and their customers tend to consider it value added to the finished system, especially since you can't just go out and buy an M1Max or an M2 for your own system with a board from elsewhere. Lots of their design and engineering is in the SoC rather than the CPU proper, too, with how it's integrated with the GPU, the memory, and the storage controller.
So in the really broad statement that only companies that make just the CPU or CPU IP add value to it, I'd consider Apple more of an edge case than an outright exception. Of course Intel, Qualcomm, Samsung, NVidia, and AMD don't build and sell only CPUs either. They make, among them, storage, RAM, display panels, GPUs, chipsets, radio baseboards, TVs, whole computers, etc. I think the point is best worded that they sell the CPUs as a retail or OEM part rather than as part of a larger system of their own. That's debatable, too, but I'm not sure Apple is a perfect counterexample to the point as I described above.
> I wonder if it's because Apple captures so much margin from the final product which means they can afford to spend top dollar on CPU design teams?
That's an excellent question. If I could answer it completely and definitively I'd probably be making my living as an industry analyst for some magazine or investment fund. I have a suspicion that might be right, though.
Apple and Google today, and sometimes other companies, are starting to remind me of AT&T before the divestiture. They're such big, profitable players that they can have huge skunkworks attacking adjacent problems to their core business and keep doing so in new directions.
> True performance RISCV designs are going to be people's money maker and never open sourced.
That turns out not to be the case.
Alibaba's C910 core -- roughly comparable to the ARM A72 cores (at the same MHz) in the Pi 4 -- is open sourced. It is being used, at 2.5 GHz, in the upcoming "Roma" laptop. That is rather expensive (for now), but I suspect the same TH1520 SoC will quickly find its way onto cheaper SBCs.
There is a very wide OoO GPL'd RISC-V core that is under development. It is aiming for eventual Apple M1 level performance. The current iteration is falling short of that at the moment, but it's already comparable to the ARM A76 in the latest RK3588 SBCs: https://github.com/MoonbaseOtago/vroom
But instead of one ISA vendor who holds all the card, there will be dozens of them, lowering prices and probably with more freedom in modifying the designs.
What you are saying isn't wrong. But I'd like to point out is that the companies that buy "Chinese RISC-V MCUs" may not be the same ones who license Arm CPU IPs and put them next to logic from other suppliers on a custom built chip.
For one thing, it is seen as risky nowadays to use Chinese IP in chips, ever since Mr. Trump said that the US government wouldn't buy certain Chinese tech.
The bigger cost in chip development is verification, not design. If you use open-source IP, I assume you need to take on this burden yourself or pay someone else to do it. The cost for doing that may well be more than the cost to license an already verified core (Arm or RISC-V ISA).
As far as I understand in Qualcomm case it is exactly a third-party implementation of licensed ARM ISA. So looks like an ideal case for switching to RISC-V.
As far as I understand, Arm ISA is a lot stricter, whereas RISC-V implementations can have more variations. So it's 'easier' to switch to a different implementation of Arm as you'll be more sure that the processor works the same way. At least, that's the idea.
We are talking here about companies such as Qualcomm or NVIDIA that have the ability to design their own cores. RISC-V is indeed completely free to them.
As long as you can get documentation and development tooling.
An MCU costing a dime is useless if all the documentation is only in Mandarin and it requires proprietary tooling including porgramming/debugging dongles that work only for one specific PC configuration. And that you need to buy/download from some obscure vendor website in China.
That's very much why you don't see Chinese parts in Western designs much. Cost per part is not that relevant if the component is impossible to do R&D with and/or support it long term. Microchip still sells even chips like 16C54 - 25 year old designs. Is the Chinese vendor going to be here in two years? And for how long are they committing to supply their parts? E.g. the first SiFive RISC-V micros are obsolete and EOL already ...
Also, the Qualcomm/ARM dust-up is not about microcontrollers (Cortex-M) but the higher performance cores. Those are licensed differently. So even if the Qualcomm claims were true (a big if!), they likely still wouldn't apply to micros.
How can this even be legal? Wouldn't this be against anti trust laws? It is like Intel saying that it is forbidden to have an Nvidia GPU in their laptop if an Intel CPU is used.
Not quite. This is more like saying what Intel has done from day 1, where a 3rd party can't take an intel CPU, add their own GPU (or whatever accelerator/silicon of choice) into the chip in place of Intel's own iGPU, and then release it, at least not without getting intel's blessing and paying a licensing fee. you can still add a 3rd party modem or gpu to an arm device, it just can't be on the same ARM SoC.
Unlike intel, ARM seems at least open to the idea to licensing such packages somewhat (see Nvidia, Apple) whereas Intel traditionally hasn't been too interested in letting 3rd parties fiddle under the lid of their silicon.
It's a regression of how things used to be with ARM, but it's not saying you can't connect 3rd party accelerators to an ARM SoC fullstop, just that it has to be seperate dies/chiplet/SoC. Your example of adding an Nvidia GPU into an intel system is still fine, so long as they are separate components.
So no, this is not antitrust worthy in the slightest.
How is that relevant? Is there a law that says you’re allowed to do that except if you have the wrong business model? Licensing the ISA does not mean that they abandon their IP or anything. They still have the exact same rights as Intel, give or take the license agreements they have in place, things like FRAND commitments, and antitrust issues.
If there is any reason to treat them differently, that would be the weight they have in the mobile CPU market, not the fact that they decided to sell licenses.
IP law is fundamentally different from product distribution law, and antitrust doesn’t really apply to IP (FRAND being a good counter-example, but that is rooted in contracts).
All IP is monopolistic, and there is no first sale doctrine. If I buy a physical book from you, it is mine to do with as I see fit. If I license the right to publish your book as part of a set, I am not necessarily allowed to sell that set into your channels and undercut your pricing. The terms of the contract dictate that, and it is not at all an antitrust violation of our contract says I am only to sell the set to schools or whatever.
> IP law is fundamentally different from product distribution law, and antitrust doesn’t really apply to IP
It is relevant when IP is used in an anticompetitive way, like Microsoft’s licensing terms back in the day. I can see a company making the argument that ARM is dominant enough that onerous terms with one company and not the others is anticompetitive (not saying that it is the case, but I have seen worse points seeing some successes recently).
Because they're licensing the design of the chips. The idea that you can impose conditions on what I'm allowed to do with the blueprints you sold me are preposterous.
Can you make me not call it an ARM chip, probably. Can you compel me not to make the chip at all -- crazy.
> Because they're licensing the design of the chips. The idea that you can impose conditions on what I'm allowed to do with the blueprints you sold me are preposterous.
It all depends on the licensing agreement. The vast majority of them have this sort of restrictions. The way it works is that a license agreement gives some rights, but not all of them. If you get a license, you have to follow the terms you agreed, which often have a whole bunch of restrictions. Whether you thing it is preposterous is not really here nor there.
It is not far fetched to have agreements covering one type of cores and not others, or things like that. Or, in software terms, a license to run some software on 12 cores and not 13. Or to install it on netbooks and not proper laptops.
A license is an authorisation to use some IP. It does not make the IP yours.
> Can you make me not call it an ARM chip, probably. Can you compel me not to make the chip at all -- crazy.
Well, then I guess the world is crazy.
In a case like this, you’d be perfectly able to make a chip, just not with the IP tied to a licensing term you don’t agree with… Well, some companies try to have more restrictive terms than that occasionally, like Microsoft who prevented OEMs from selling computers without Windows, or Google who prevents OEMs from making Android-derivative devices.
tl;dr if you put it in the contract and you agree to it then it must be fine.
This kind of thinking makes the world worse for everyone except parties that have leverage. No matter what, no company should be able exert so much will over another that it actively prevents anyone but them iterating on their design.
Any time you see an agreement like “we’ll give you x only on the condition you not compete with us” alarm bells should ring.
While the background makes sense, I don't think the conclusion follows (that this isn't an antitrust issue in the slightest).
Frankly - many of those other firms are antitrust worries too; it's just that legal antitrust tools are quite weak and (IMHO) insufficient for ensuring competition, and other borderline monopolies likely have better lobbyists. And that might work against ARM here, because the very customers ARM will be squeezing here are those entities with potentially harmful amounts of market (and political) influence.
We'll see how this develops, but I wouldn't take it for granted that this is not antitrust worthy in the slightest. It probably won't get to that, but to entirely exclude the possibility...
Antitrust law generally applies to those with large or dominant market positions, to keep them from distorting the market in order to protect their position.
ARM is certainly large but I don’t think anyone would argue there are no alternatives (this thread is full of many!).
And the angle to get antitrust here is pretty nuclear: it would require saying that Arm IP is so critical that the company has lost the right to license it as they see fit. Not quite a revocation of their IP rights, but close.
So yeah we can’t totally rule it out, but such an action seems incredibly unlikely.
Especially given the specifics of the complaint here, where someone wants to leverage a beneficial price they get from Arm to undercut Arm in the market. It reminds me of that “TV antennas in a datacenter” company that sought to leverage broadcast TV licensing for internet streaming delivery: a clever troll but never going to stand up to contract law, and unlikely to get regulatory support.
Aren't they saying it has to be on a different chiplet? It's not like they can't have them "on the same laptop", AFAIk they're on separate chiplets or even chips today.
That’s not what it means, they mean on same chip and equivalent would be, Intel CPU and Intel on board graphics on its CPU.
It would be a suicide move if ARM pulls off what you said, forget corporations, consumers will revolt.
At this point it appears as if ARM is run by former Nokia Kodak and Yahoo executives.
For similarly innovative tech strategies I suggest the book “In search of stupidity” by Merrill Chapman
There is a lot of misinformation being spread out.
Arm changed the business model from requiring the SoC vendor to license every single piece of IP individually up-front (one license for CPU, one license for GPU, one license for interconnect, etc) to a "all you can eat model" and "pay only once you shipped the product":
This is very positive because it simplifies the legal paperwork for licensing IP (one license gives you access to an entire portfolio of IP instead of just one) and lowers the bar for prototyping because you won't be required to pay full cost for a license you don't you if you will use. Qualcomm is heavily misrepresenting this to their light.
oh come on, the majority of people in this HN post are taking it as the state of things now.
Very few posts are even mentioning that it’s Qualcomm claiming it.
This is a classic misinformation tactic. To present one sides claim as fact and then by the time the truth comes out one way or another, it doesn’t matter because the sensationalist reporting is what people remember.
I think it’s because there’s sooooo many software engineers or sys admins on here who work on higher level applications.
Anytime something lower level comes up (like graphics APIs or OS architecture) or hardware related things, the quality of comments drop because there’s less familiarity (and perhaps an over abundance of confidence)
I forsee RISC-V ports of Android rapidly gathering speed...
Ports of Android 12 are already running on dev boards. Most of the work was done by Alibaba, and Google has just agreed to merge it into AOSP, making RISC-V a supported Android platform.
If they do, all that's needed is a good compat layer that can emulate ARM for app compatibility, and the app ecosystem to be updated to compile all the native code to RISC-V for new apps too. Intel tried and mostly failed at this step, with Googles help, so it's a tough task.
Remember most native libraries are native libraries because they do some compute-heavy task. They probably involve lots of hand crafted SIMD ARM assembly. Porting that to RISC-V isn't trivial, and most app developers can't be bothered for the tiny number of low-end RISC-V users.
Instead, those users end up using the ARM apps via an emulator, and that makes them even slower on an already slow cheap phone.
End result: RISC-V gets a reputation as a very slow and very crap CPU architecture. Just like MIPS and x86 on mobile got.
NEON isn’t a great SIMD design and is rather old. Everyone’s already needing to start porting to SVE for ARMv9 anyway. At that point, porting to another ISA at the same time isn’t a huge deal as you’ you’re already in the code.
If you put RISCV and ARM assembly next to each other, you’ll see that RISCV is much easier for humans. The vector extension is also much easier to understand than SVE.
I believe porting old NEON too RISCV vectors will be easier than porting to SVE.
Apple transitioned from x86 to ARM in less than a year.
It can be done, using mix of native applications and emulation. And as long the emulation speed is indistinguishable from native (Like Rosetta 2), end users won't notice it and won't complain.
Google & Meta can team up to transition all their apps to RISC-V and that would be 70% of the most used daily apps, other will follow the lead.
I think they'd been working on it for many years beforehand. Getting the software stack ready for a new architecture, especially one where there is a lot of 3rd party code you can't touch, is a lot of work.
After all, step one was porting their darwin kernel in 2007. Then the 'Unified Platform' work of 2017 made the userspace work on ARM... Then 2020 they released an ARM desktop product... So it was a lot of years!!
I think you'd be shocked how much of my NixOS system not only natively compiles for my Vision Five but also cross compiles to it... (and Nixpkgs cross compiling support has room to improve, so..)
Obviously a monorepo of largely FOSS is easier to deal with than random Android apps but I don't think it's that dire or insurmountable.
> And as long the emulation speed is indistinguishable from native (Like Rosetta 2), end users won't notice it and won't complain.
Apple could do Rosetta 1 and 2 because the emulated CPU was considerably slower the new architecture. No evidence at all that this would be the case here.
Well. QuickTransit was from Transitive Corporation, which was acquired by and then demolished by IBM, a bunch of the people from there are now at Apple, for one. The ideas are pretty straightforward and many of them show up in JITting emulators and high-level emulation approaches. I haven't seen an analysis of Rosetta 2 but I imagine it is pretty similar.
Transmeta is different - Transmeta Crusoe etc. were a weird VLIW-like ISA plus their ISA-specific code morpher sw.
Apple transitioned from x86 to ARM in less than a year.
Apple started a lot earlier than the announcement and hasn't completed the transition yet (Mac Pro). A lot of companies better have had RISC-V initiatives in the pipeline already.
Packages consist of a bunch of java-like bytecode, and some native code. Usually the native code is for video decode libraries, neural nets, openssl, javascript engines, speech recognition stuff etc.
The bytecode is platform agnostic, but the native code isn't.
When the app is installed or run for the first time, the bytecode is compiled on device and will run fine. But the native code from the package will have to be emulated if the app is to run. Since the native code might want to link with system libraries (libc etc), there needs to be a rather complex shim to keep everything working properly.
ARM is likely well geared for managing and collecting royalties from chip vendors (hundreds?). It seems to me that setting up an organization to collect from actual device manufacturers (~10^6-10^7 or more?) is a completely different scalability scope and, therefore, a completely different business. Maybe this is rather a targeted game of chicken tailor made against a specific licensee?
Also, re RISC-V. I'm all for that. But an ISA does not make a chip. There is far more to what ARM offers than just its ISA: production-grade cores and surrounding blocks on bleeding edge nodes, etc. They could, theoretically, start selling a RISC-V fetch and decode unit (think of it as a Unix API on top of Windows) and still come out on top.
Collecting from actual device manufacturers would be beyond crazy, fully agree about that. But it would be almost trivial to do for some core classes, at the upper end, if they really wanted to. Certainly even that would still be hurting themselves more than others.
Assuming for the moment this is real, could somebody ELI5 how this would work in practice in the face of the first sale doctrine?
(1) Chip manufacturers can't use ARM IP without appropriate permission/licensing in the first place, so they would _still_ need some ARM contract to even make the things, right?
(2) The first-sale doctrine doesn't grant unlimited rights -- OEMs can't reverse engineer and clone a chip -- but certainly they can attach it unadultered to other components so long as there aren't separate patent issues for that particular combination of underlying technologies? Does ARM perhaps have a patent on stuff that looks like "ARM Core" + "Neural Accelerator" or something like that?
(3) Assuming there aren't any loopholes out of first-sale, is the play that in the contract in (1) ARM prohibits direct sale and requires chip manufacturers to profit via some contract that includes language about OEMs having to pay ARM? Is that sort of law-skirting anti-competitiveness actually allowed here or in other domains?
The only thing this will result in is the speedup of the move over to RISC-V. Because of this spat I predict RISC-V will be dominant within a decade and large inroads in 5 years or less in the mobile and embedded space.
I even predict we'll see large numbers of vendors who'll make RISC-V server and desktop CPU's.
adoption of ARM just started to pick up in server & laptop market, it could be a nice alternative to x86, yet short sighted decisions like this from ARM is going to cause huge uncertainties.
RISC-V is great, I am all for it, but let's be realistic, many people including myself have been waiting for 2 decades to see the current upward trend of ARM in servers & laptops, how many extra years we will have to wait for RISC-V to catch up and get to ARM's current position? An extra 5-10 years? Intel must be laughing!
As an ex-ARM & ex-Qualcomm employee - middle finger for both companies, you greedy licensing bitch.
5-10 years isn't that long. I guess it depends on the management philosophy, but if someone told me I would my company would be losing a large market share in 5-10 years I would be panicking.
This is what, I believe, Nvidia wanted to do. It may be a dick move but they want to make a ton of money, damn it. Even if it drives many of their customers to risc-v
Arm had a good idea to drive up sales of their non-core IP, didn't realize how much device manufacturers really like device specialization for sub-ms asynchronous time warp, low power hot-word detection, not needing to get directly involved in fab, nor did they realize how much this looks like an antitrust case since "i wannabe the sole chip manufacturer on the board" isn't that subtle, "chips per board" is actually a reported metric in quarterlies for public chip companies.
https://www.semianalysis.com/p/arm-changes-business-model-oe...
HN submission: https://news.ycombinator.com/item?id=33371725