ARM Co-Founder Hermann Hauser: 'It's In Nvidia's Interests To Destroy Arm' (newstatesman.com) 87
An anonymous reader quotes a report from NS Tech: SoftBank is in advanced talks with US chip company Nvidia to sell Arm -- with a price in the region of 32 billion euros reportedly being thrown around. But Nvidia's purchase of the Cambridge-based chip designer would not only strike a blow to the UK's technological sovereignty, but would result in the destruction of Arm itself, Arm co-founder Hermann Hauser has claimed. Nvidia recently overtook Intel as the most valuable microprocessor company in the world, and its great wealth right now provides it a unique opportunity, says Hauser. "They are the semiconductor company that can buy Arm to destroy it -- and it is very much in its interest to destroy Arm because they [would] gain a lot more than the 40 billion that they pay for it," he claims.
Hauser says the acquisition would allow Nvidia to swipe "the microprocessor crown from Intel", and become the chip supplier for 95 per cent of mobile phones, 90 per cent of embedded controllers for the internet of things, as well as taking the PC and data centre markets. Hauser says that an Arm acquisition is a means for Nvidia to become the dominant microprocessor company in the world, "at the same time as they can prevent their main competitors" from making use of further Arm developments. Instead, competitors would need to "scramble" to create their own architecture, handing Nvidia the edge. Up until now, Arm's business model is acting as the "Switzerland of the semiconductor industry" -- maintaining relations with many customers around the world. "Most of them of course, are competitors of Nvidia," says Hauser. Japanese technology investment firm SoftBank acquired Arm in 2016, but ensured the company's continued neutrality.
Hauser says the acquisition would allow Nvidia to swipe "the microprocessor crown from Intel", and become the chip supplier for 95 per cent of mobile phones, 90 per cent of embedded controllers for the internet of things, as well as taking the PC and data centre markets. Hauser says that an Arm acquisition is a means for Nvidia to become the dominant microprocessor company in the world, "at the same time as they can prevent their main competitors" from making use of further Arm developments. Instead, competitors would need to "scramble" to create their own architecture, handing Nvidia the edge. Up until now, Arm's business model is acting as the "Switzerland of the semiconductor industry" -- maintaining relations with many customers around the world. "Most of them of course, are competitors of Nvidia," says Hauser. Japanese technology investment firm SoftBank acquired Arm in 2016, but ensured the company's continued neutrality.
Re: (Score:1)
It's only a monopoly if they manage not to kill it in the process. Apple and a few others purchased licenses to build their own, so they can continue designing and selling ARM-likes even if nVIDIA stops future sales. nVIDIA also isn't exactly known for making huge progressive changes necessary to supply the mobile market and won't out-compete either AMD or Intel in the future on the desktop.
nVidia is the Intel for GPU's, they are the overall best performers, but not necessarily the least power hungry. Their
Re: (Score:2)
nVidia is the Intel for GPU's, they are the overall best performers, but not necessarily the least power hungry. Their 3080 will consume something like 300W, their mobile GPU lunch has been eaten by ARM and their embedded market GPGPU by both ARM and Intel.
So one of these Samsung phones with really big battery would run about 40 seconds until the battery is empty. And then it would melt :-)
Re: (Score:1)
Isn't that what Samsung phones do already?
So much for Apple Silicon (Score:1)
Comment removed (Score:5, Insightful)
Re:So much for Apple Silicon (Score:5, Insightful)
Do you have evidence that Apple's ARM license is not unlimited and non-revocable? I have a hard time thinking that Apple would have committed its future to an entity it did not control.
To me this is more likely to lead the the end of ARM, and of those who need ARM-like processors licensing Apple's previous generation A-series chip.
Re:So much for Apple Silicon (Score:4, Informative)
I'm sure Apple's license for current ARM products is pretty airtight.
Not sure about their license for future ARM products though, because there is never really any guarantee there will be any or that you will even like them if there are.
Re:So much for Apple Silicon (Score:5, Informative)
The last major Apple chip that used an ARM product was the A5 in 2011. It featured the ARM Cortex-A9 processor design. Everything after that was custom in-house designs leveraging little more than the instruction set specifications. Apple's license (on at least what they currently use) doesn't expire and can't be revoked, so they don't really care what happens to ARM. They don't need to maintain binary compatibility with any platform but their own, so if they want to, they can simply make their own changes to the instruction set as required.
Re: (Score:2)
If Apple is only using legacy ARM IP then I agree. No worries if they go away.
Re: (Score:3)
That is not true. For example Apple used ARM's big.LITTLE architecture and is likely to move to the replacement whose name I forget too.
ARM did a huge amount of work getting ARM cores to perform well in relatively large groups while sharing resources such as RAM and remaining low power. Apple made use of that work, as did everyone else making high end big.LITTLE architecture SoCs.
Re: (Score:2)
That is not true. For example Apple used ARM's big.LITTLE architecture and is likely to move to the replacement whose name I forget too.
Apple took the idea that you could have a chip with some powerful and some less powerful processors. They don't need a license for that. And what's in a current iPhone is most definitely not ARM's big/little design but something much more advanced.
Re: (Score:2)
Not just the idea, the entire big.LITTLE architecture. Apple's design is an implementation of big.LITTLE.
Up to the A10 they couldn't use both the high performance and energy efficient cores at the same time, it was one or the other. That was a common feature of early big.LITTLE designs and Apple actually lagged a bit keeping up with ARM's improvements that allowed all cores to run at the same time. The A11 was the first to adopt the updated big.LITTLE architecture, about a year behind the competition.
So not
Re:So much for Apple Silicon (Score:5, Funny)
I think they will stick to a specific ARM ISA and look down on anyone adopting any future changes. In time, they will be known as the Armish.
Re: (Score:2)
There is a mod box there. Would for sure get a funny from me but I think it is a bug.
Re:So much for Apple Silicon (Score:5, Informative)
Apple has a permanent and irrevocable 'architecture' license to Arm at least through ARMv8.6-A. So if nVidia were to buy Arm, they would be unable to stop Apple from continuing to develop derivatives of the existing ARM8 family. So worst case scenario, you would be looking at something like a fork where Apple's Ax chips and future ARMv9 chips diverge. Apple could easily keep ARM8 chips competitive for 5+ years while consider future options
nVidia has plenty of other reasons to buy Arm, but screwing Apple over isn't going to be one of them.
Re: (Score:1)
So Apple can continue to fork ARM.
Their architecture will continue to grow proprietary. It's how Apple rolls. The 'open' parts of their OS have tended to wither.
Re:So much for Apple Silicon (Score:5, Informative)
Just to expand on this, Apple essentially forked Arm back in 2012, with the A6 chip. Since then Apple has been doing their own designs that share a instruction set with ARM (with some additions) but they haven't used any of the Arm Cortex designs for 8 years. That part of why Apple is so anal about calling their new processors 'Apple Silicon' instead of Arm because, well, they aren't Arm chips.
Re: (Score:2)
The age of arm architecture and any associated patents are what it is all about. Leveraging that for as much as possible before it expires ie Nvidia have become really greedy and are likely looking to extort as much as possible and fuck future business, extort everything they can now, Now, NOW.
Locking up the market and hugely inflating prices.
Re: (Score:3)
Apple would have to invest some serious cash to keep their forked parts competitive. For example ARM is now pushing DynamIQ which is the replacement for big.LITTLE, something Apple made extensive use of in its CPUs. So if Apple got locked out of DynamIQ IP they would have to develop their own to stay competitive, and keep in mind they are already way behind on that kind of multi-core performance as you can see in benchmarks against the Snapdragon 865.
There may be other issues like proprietary busses for per
Re: (Score:3)
Nvidia hasn't even announced a binding offer for ARM. You realize that Apple is ~6x the size of Nvidia and could outcompete Nvidia for ARM if it needed to...
Re:So much for Apple Silicon (Score:5, Insightful)
Chance of regulators allowing Apple to purchase primary designer of chips for competitors of their most profitable sector is between zero and "and while at it, we're going to look at other parts of your business for signs of illegal monopolistic behaviour looking to break you up".
There's very little chance that Apple will want to even pretend like they want ARM. That is a one can of worms that actually can screw their entire business over in a major way. There's just no reason for them to go there.
Re: (Score:2)
Chance of regulators allowing Apple to purchase primary designer of chips for competitors of their most profitable sector is between zero and "and while at it, we're going to look at other parts of your business for signs of illegal monopolistic behaviour looking to break you up".
Where have you been the last 3 years?
Re: (Score:2)
Re: (Score:2)
I can say the exact same thing about various Trusts.
And they went away.
Re: (Score:3)
But then you notice all your friends are standing on the same rug. There's about a zillion micro-controllers based on ARM as well, pretty much all of the popular 32 bit ones. Mostly cortex-M0 and cortex-M4. The larger SBCs including Beaglebone and Raspberry PI are ARM as well.
We're at the point where this sort of thing probably shouldn't be entrusted to capitalists who would gleefully set the world on fire if the pay was right.
Re: (Score:2)
Stuff like Beaglebone that is based on TI processors don't need to worry at all, TI isn't going away and neither are their ARM processors.
And if TI wanted to move to RISC-V or something, they could make the switch transparent to users because of their mature and easy to port driverlib ecosystem of C APIs.
Raspberry PI is Broadcom, they also have nothing to worry about. These companies already have their licenses, and they only licensed an instruction set; the actual processors are of their own designs.
Re: (Score:2)
The current products will stay on the market, of course. But the chipsets they use become instant dead-ends, EOL for the entire product line. As do all the other ARM based chips out there.
It wouldn't be the death of the industry or anything, but it would make everything more expensive and less innovative for a fair while. A metric assload of wasted effort.
Re: (Score:2)
Nvidia's market cap is about £300 billion. Apple has nearly $200 billion in cash. If Nvidia ever threatened Apple's use of ARM, Apple would have little trouble raising the money to launch a takeover bid.
It' much more likely though that, if Apple's licence agreement turns out not to be airtight, Apple would just pay Nvidia some money to carry on using ARM.
Re: (Score:2)
There's schadenfreude in knowing that Nvidia pulled the rug from underneath Apple. Given how Apple just gave Intel the finger, now their bridge to ARM looks to lead to nowhere without a future.
Depending on how Nvidia plays their hand, the rest of the industry could make the job to RISC-V. But at the moment, doing that now is a bit (wait for it)...risky.
Or Apple, with it's own in-house chip design staff, could simply fork the ARM architecture just enough to keep them from getting sued, and proceed with the plan to completely bring everything under Apple's own roof anyway. Recall that Loongson did something similar with their MIPS clone chips awhile back.
Re: (Score:2)
As I understand it, Apple holds a priority license to ARM from its original formation; they would not easily be harmed by the acquisition. Other companies are not in the same position.
Nvidia doesn’t need to invest anything in ARM, and can let it die off that way, but they cannot just halt all existing licensees or modify terms.
Re: (Score:2)
It may turn out to be hard to make an instruction set die by neglect, as new microcontroller models generally are sold based on having improved peripherals, rather than new instructions.
And you can add your own new instructions to your design if they won't license the new models to you, or if they stop making new models. It may turn out that you don't actually need any new forms of addition, subtraction, or register access to implement the new features actually being added to products.
Re:So much for Apple Silicon (Score:4, Informative)
Re: (Score:2)
Sale will be conditional (Score:2)
They need approval of every government that regulates mergers. Obviously these concerns will be among the conditions for that approval.
Re: (Score:3)
That's the US and Japan, in this case.
Japan isn't going to tell Softbank it can't sell some UK company it bought.
And the US is not likely to be worried about a US company buying a formerly British company from Japan.
That's the approval they need. The US Government. Not "every government."
What if? (Score:2)
Re:What if? (Score:4, Interesting)
Adopt RISC-V most likely. Getting mass fabrication of RISC-V chips would still be a problem, but at least it's an open platform, so licensing goes out the window. Frankly, I think it's likely in the industry's very best interest to start working towards this ASAP. It's clear there are a whole lot of eggs in the ARM basket.
Re: (Score:3, Interesting)
Or PowerPC. Same licensing reasons, as of a couple of years ago (though I'm not sure Apple would have had to pay licensing fees for their own PPC designs anyway, given that they were one of the founding companies that built the architecture).
Re: (Score:3)
IBM "built" the entire PPC architecture. All of the reference designs were done entirely by IBM and the 601 was literally a converted Power processor (and not fully PPC compliant).
Re: (Score:3)
While it's true that the PowerPC design is loosely derived from Power 2 (with lots of stuff removed and added), the stuff that was added was at least in part designed by the Apple and Motorola teams. In particular, I'm pretty sure that the Altivec instruction set was largely driven by Apple's side.
Either way, the instruction set is apparently freely licensable now, so it's kind of moot. :-)
Re: (Score:2)
Doubtful. RISC-V is a mess. Until someone goes in and reduces the current "federation of standards" to a few ISA standards that the industry can agree upon, there's the risk of vendor ISA extension hell:
https://en.wikipedia.org/wiki/... [wikipedia.org]
Many of the extensions are still subject to change, and there's no particular requirement for vendors to include most or all of what you would expect in designs that would/could replace ARM.
Apple isn't the only company out there with a design license for ARMv8, and even tho
Re: (Score:2)
Just look at how ALPHA was handled after Digital was bought up
Intel was a strong partner with Compaq and Compaq licensed ALPHA to Intel
Intel was a strong partner with HP and demanded that ALPHA chip benchmarks be withheld while HP sold Itanium servers
Eventually, much of the ALPHA team joined Intel resulting in Itanium II
Key difference between then and now
Intel had a product that was a direct (although woefully inferior) competitor to ALPHA, whereas nVidia has a huge hole in their portfolio where ARM fits ni
Re: (Score:2)
So what would happen if nVidia did buy ARM and decided to disband the company, charge extortionate licensing costs, or use selective licensing to pick and choose who gets to use ARM designs? How would the industry react to such hostage-taking behavior? Wouldn't that be simply inviting regulatory trouble...and if so to what end?
It would have no visible effect at all. ARM doesn't design or license ICs. They license an instruction set. Existing customers already have contracts for the existing instruction sets, and new instruction sets are not something normally needed. All that is needed is instruction sets for the different classes of processors, from small to large. New ARM licenses aren't involved in the design of new "ARM" microcontrollers. Only when a company is designing a processor from a particular class for the first time
Obligatory... (Score:1)
Would MS buy it? (Score:1)
Re: (Score:1)
would apple buy it?
Re: (Score:2)
If it's that dangerous... (Score:2)
If selling ARM is truly such a danger to companies using it, why wouldn't someone like Apple come in and double the offer just to keep ARM safe?
Surely they are in the best position to judge what kind of exposure this sale presents to them.
Re: (Score:2)
For 32 billion euro? 38 billion USD at today's rates. For that money Apple could design their own architecture from scratch and build its own fabs if it wanted to.
Which tells us Apple is not concerned with who controls the ARM entity and what their license terms for that architecture must be.
Wow - maybe Apple could buy the PDP-10 architecture from whoever owns it now, upgrade it to 72-bit, and run with that!
Re: (Score:2)
Re: (Score:2)
> buy the PDP-10 architecture from whoever owns it now
DEC was sold to Compaq, that merged with HP. The rights should be with HPE now. I'm not sure they would license it to Apple.
But's it's irrelevant, Apple can simply move to RiscV or PPC if they need to.
Re: (Score:1)
Wow - maybe Apple could buy the PDP-10 architecture from whoever owns it now, upgrade it to 72-bit, and run with that!
That's a kool idea! I support it and would even buy an iphone that runs TOPS-10 in a VM
Re: (Score:3)
Apple would buy a heap of headaches if they bought Arm. For one, it would create a mountain of antitrust issues surrounding Apple having their own proprietary Arm architecture chips while simultaneously licensing reference designs to virtually every smartphone manufacturer out there. Apple could avoid that by making their proprietary Ax chips available to as Cortex designs, but that defeats the whole purpose of developing them in the first place. And still leaves issues over the fact that Apple openly st
Re: (Score:2)
Because Apple doesn't use ARM chip designs and has an irrevocable license to the instruction set (the only part they use). They don't care if ARM falls into a black hole tomorrow. They'd just make their own instruction set extensions.
Re: (Score:1)
That's kind of what I was thinking, which is why I was saying that if the sale of ARM were actually an issue whatever company it was an issue for would buy ARM instead... since Apple is not buying, it must mean as you say there is no danger to Apple at least.
Re: (Score:2)
If selling ARM is truly such a danger to companies using it, why wouldn't someone like Apple come in and double the offer just to keep ARM safe?
They wouldn't double the offer because there is no need to increase it by more than one percent. And Apple wouldn't go in anyway because they are not under threat by this, and it would be a legal nightmare if Apple controlled the chips that all its competitors need.
Apple might join a consortium, if Samsung pays $10 billion, and Qualcomm pays $10 billion, then Apple might add another $10 billion. And then bid it up just to make it expensive for Nvidia.
And I could imagine that Apple's license allows the
RISC V (Score:3)
Good .. this may allow a competing ISA. Hopefully a fully open and royalty free architecture such as RISC V can come up and take it over.
Re: (Score:2)
I've been waiting until some decent RISC-V silicon comes to market to try. Most of my embedded systems are quite traditional. Lots of them run Linux and the majority of the code is quite portable C code. Retooling to a different architecture would not be all that painful.
Hopefully just the rumblings of this will show some of th
Re: (Score:2)
For RISC V to be like ARM is now it needs a big backer willing to put very large sums of money in to development. ARM isn't just a good ISA, in fact it's not /that/ great, but they have an army of engineers improving it and supporting companies that use it.
Say you want to build a RISC V SoC with 4x high performance cores and 6x power efficient cores, similar to a high end ARM SoC. Are there any advanced, fully qualified and ready-to-go designs for that sort of thing? Can you see actual real-world performanc
Re: (Score:2)
Good .. this may allow a competing ISA. Hopefully a fully open and royalty free architecture such as RISC V can come up and take it over.
People don't care about fully open and royalty free, they care about high performing and cheap. And for that you need to get the top chip designers. Which RISC V doesn't have.
Nvidia won't kill ARM (Score:5, Insightful)
NVidia has GPUs for AI and fast end-to-end networking after acquiring Mellanox. The major things they are missing are memory modules and a CPU product. Memory is a commodity product available from a number of manufacturers, so they don't need to develop that expertise.
But Nvidia NEEDS in-house CPU products to be able to create a complete super computer without going to Intel or AMD. That's why Nvidia wants ARM. They won't destroy ARM. They will use the company to develop CPUs capable of being used in exa-scale systems.
Nvidia would probably be happy to let the current ARM licensees have the products they want on their current terms. The company is primarily interested in development of future CPUs to fit their perceived needs.
Re: (Score:1)
I like this take. ARM CPUs made it to the Top500 after all.
Re:Nvidia won't kill ARM (Score:4, Interesting)
It seems uneconomic for Nvidia to pay $32B to get an exascale processor development team.
However, Hauser does have a point. Google has gradually morphed Android from open to proprietary, an example Nvidia has surely studied cafefully.
A gradual introduction of Super Arm features by Nvidia might well allow them to shift the Arm ecosystem towards a more proprietary structure with Intel like margins for Nvidia.
Re: Nvidia won't kill ARM (Score:1)
The problem with Nvidia is, they simply cant help themselves.
Every single company that has partnered with them has run away extremely pissed.
They remind me of the scorpion fable.
Re: (Score:2)
It seems uneconomic for Nvidia to pay $32B to get an exascale processor development team.
The Nvidia Tegra Pascal and Xavier chips both have custom ARM cores, so I assume that Nvidia already has a CPU architecture team. Even so, it's the GPU and not so much the CPU that is the driver for exascale. It's not clear what Nvidia gets from ARM that it doesn't already have.
Re: (Score:2)
Wouldn't it be much cheaper to just license ARM tech like everyone else does for that?
Re: (Score:2)
If they bake NVLink into ARM reference server designs (think Neoverse), it could help them out immensely.
Re: (Score:2)
ARM license allows you to do whatever you want with the designs. Again, no purchase of ARM necessary.
Re: (Score:2)
Existing licenses do . . .
Re: Nvidia won't kill ARM (Score:2)
If they need ARM CPUs to get that, they do not need to buy ARM, they can simply license their designs (as they already do, btw).
Re: (Score:2)
Re: (Score:2)
Thank you for stating the obvious! nVidia doesn't want to be deplatformed. Period! They aren't a "microprocessor" rival to Intel. They sell compute cards and video accelerators, and that's what they want to keep doing.
Micro Men (Score:5, Informative)
Oh and this should be a wake-up call to get more hands working on RISC-V.
Ramping up the price? (Score:1)
Self proclaimed mobile CPU marker monopolist argues that others will do worse than him.
What about free market? (Score:1)
Re: (Score:2)
There's no such thing as a "free market"
The market is always controlled by some player. You can allow it to be regulated by a democratically elected government that represents the people. Or you can let it be controlled by huge hedge funds that have only their self-interest in mind.
There's always a group of people that make that playground rules.
The myth of the "free market" is intended to let the rich/powerful close the door on any emerging competition.
That's why we still use polluting oil and coal. Why th
Most of them are competitors? (Score:2)
Mostly the ones using their own GPU are competitors and then in the loosest sense. It's not their core business, if NVIDIA offers a better GPU at a reasonable price they'd gladly use it. All the companies using Mali obviously aren't competitors, they'd have to switch from one third party GPU to another ... big woop.
Maybe Fujitsu is a competitor for their supercomputer IC, but that's small time.
Makes sense (Score:1)
This would really make sense for Nvidia. I don't think they'd kill it, they'd embrace it. Maybe we can finally put the X86 architecture in history museum.
What about embedded systems? (Score:3)
All the comments I have read have focused on either the desktop/laptop market or mobile processors. But there is another realm where ARM is very important: embedded systems. ARM's Cortex-M series is coming to dominate the market for microcontrollers; they pretty much rule 32-bit controllers, and ARM-based parts are now cheap enough that they are starting to displace 8-bit microcontrollers as well. Many manufacturers make them, including Atmel Microchip, ST Microelectronics, NXP, and Freescale.
In the microcontroller space, the manufacturers use the reference Cortex designs with few changes. The changes they do make are mostly in access to flash memory, and for the faster parts to high speed on-chip RAM. The peripherals, on the other hand, are completely different from one maker to another.
In the long term, the industry would probably coalesce around another architecture such as RISC-V. But the disruption to that market in the short to medium term could be severe if NVIDIA were to buy ARM and not continue development and friendly licensing of that series of designs.