ARM helped Nuvia develop a core for servers where ARM wanted more marketshare. Qualcomm buys Nuvia and decides to put the Nuvia core in laptops and phones first.
Now of course, ARM is not happy about that. They spent resources to help Nuvia develop the core for servers. Now Qualcomm is using the Nuvia core to compete with stock ARM Cortex designs in phones, laptops, and servers.
ARM makes money from ISA and stock-ARM core licensing. They make way more money if you use their stock cores than if you only use their ISA. Oryon would only be using the ISA. Apple SoCs also use only the ISA.
Not only does ARM lose money from Qualcomm switching from Cortex cores to Oryon cores, they now have a legitimate competitor for Cortex SoCs other than Apple.
Qualcomm was also the the strongest lobbyist against the Nvidia/ARM deal. They don't want the competition from stock ARM cores backed by Nvidia. They don't want to compete with a combined Nvidia+ARM. They just want to compete with stock ARM cores with their Nuvia cores.
Other ARM vendors like Mediatek, Samsung have no ability to create better cores than Nuvia's cores. They can only rely on stock ARM core designs. Therefore, Qualcomm's Nuvia-derived cores should always be the high end* while other ARM vendors can only fight in the mid-range, which have much lower margins. Had Nvidia acquired ARM, Qualcomm would have been competing against ARM + Nvidia IP + Nvidia money.
I actually bought some Qualcomm stocks because I saw how Qualcomm outmaneuvered ARM.
*Assuming that stock ARM cores won't surpass Qualcomm's Nuvia-derived cores
I have not heard anything about ARM helping Nuvia in any way.
The legal complaint of ARM has been that the license they had granted to Nuvia was limited in scope, i.e. it allowed Nuvia to only design a core for server CPUs, so as soon as Nuvia has repurposed that core, the existing ARM license has been lawfully terminated.
Moreover, ARM has argued that even if the buyer of Nuvia, Qualcomm, has an ARM license for CPUs that can be used in laptops, this license is not applicable to the Nuvia cores, because a significant part of their design has already happened under the Nuvia license, so it has become illegal after that license had been terminated.
So ARM wants that Qualcomm should either buy a new, more expensive license, or destroy all the design work leading to the Nuvia cores and start a new cleanroom design of laptop cores, to be used perhaps in their 2029 laptop products.
Without access to the text of the secret contracts between ARM and Qualcomm and between ARM and Nuvia, it is impossible to decide which of the legal arguments of Qualcomm or of ARM are a complete BS.
Normally such contracts should have been unambiguous enough that such a legal fight should have never been possible, but it seems that at least one of the parties has concluded that it is possible to twist the meaning of the contracts according to their business interests.
In any case, conditioning a design license in function of the application ___domain seems abusive. What could be acceptable would be only a per-device royalty fee that is computed differently depending on the application ___domain.
Even if ARM's legal reasoning has a relatively low chance of success, I suspect the likely difference in the value of the contracts is enough that it's worth them trying anyway, either to cause enough of a threat to extract some kind of settlement from Qualcomm or for a chance of a really big payout. It's probably not a great long-term plan: such actions are likely to hasten work on RISC-V alternatives at Qualcomm and all the other manufacturers witnessing the attempt.
Would it be any better for the whole market if Qualcomm developed their own proprietary RISC-V cores with proprietary/patented extensions that most software would be reliant on? There wouldn't be any equivalent of ARM to keep them grounded and they'd have zero incentives to give away their IP for free to their competitors (or even license it).
It's probably not a great long-term plan: such actions are likely to hasten work on RISC-V alternatives at Qualcomm and all the other manufacturers witnessing the attempt.
I think any Qualcomm work on RISC-V is mostly for show. It's an empty threat. Qualcomm can't move to RISC-V because they don't have the power to drag an entire industry to optimize software for RISC-V.
Qualcomm does not have any pull to make Google to make RISC-V Android a first class citizen. Nor do they have any pull to make Microsoft develop Windows RISC-V. They also don't have a presence in servers so they can't make Linux and server software like Kubernetes/Postgres/Redis/etc. optimize for RISC-V.
The actual entities that have this kind of power is Apple, AWS, Microsoft, and Google and some Chinese tech giants.
The licenses provided Nuvia access to specific Arm architecture, designs, intellectual property, and support in exchange for payment of licensing fees and royalties on future server products that include processor cores based on Arm’s architecture, designs, or related intellectual property.
Arm provided preferential support for Nuvia’s development efforts, with Arm seeking to accelerate research and development in next-generation processors for data center servers to support that sector’s transition to Arm technology.
The paragraph about "support" is void of any content.
Arm does indeed provide significant technical support to the companies that buy Arm IP, as it is normal for any vendor, but that does not apply to companies like Nuvia.
For companies which buy ISA licences, there is very little need for technical support, at most some clarifications about how to implement the compliance tests. There are only a handful of companies with ISA licenses, which have only very infrequently needs for technical support, so even if a single Arm employee is assigned for this task there is no need for "preferential support" in order to respond immediately to any support request.
"Arm seeking to accelerate research and development in next-generation processors for data center servers to support that sector’s transition to Arm technology" just means that Arm has continued to act as a competitor for Nuvia, in developing their own Neoverse server cores and related infrastructure.
It is true however that only because the Arm CPUs have a minor share of the server market for now, the progress for one of them can still have a synergetic effect on the others, instead of an antagonistic effect, so there is a little truth in the claim of Arm that they have worked "to support that sector’s transition to Arm technology", but in any case that does not mean that Arm has spent any cent with the support of Nuvia.
All the text that is public about the Arm-Qualcomm conflict is useless, because it avoids to quote the relevant paragraphs of the contracts between Arm and both companies, which would have clarified instantly which party is right.
Anyone knows where the claim that the Nuvia licence was server-specifc comes from? This seems to be a bad copy-pasta from the Reuters article (which does not claim this), and no company ever claimed this AFAIK.
Edit: the answer can be found in the court documents, here Qualcomm's Sept 22 answers to Arm's claims: https://storage.courtlistener.com/recap/gov.uscourts.ded.798... ; the gist is that Arm claims Nuvia's licencing fee was higher than Qualcomm's because it was negotiated with the intent of making server chips. Qualcomm admits as much:
> Defendants further admit that NUVIA and ARM intended the licensing fees and royalties set forth in the NUVIA ALA to apply to future server products, not products for other markets
How does ARM license the ISA? Isn’t this cloneable like the shape of an API? Intel never permitted the 586 pentium clones. Why does ARM get this veto if the cores are developed from scratch?
I understand that in the Nuvia case they were not developed from scratch. I just don’t understand why Apple and Qualcomm don’t just remove ARM licensed tech from their chips and maintain ISA compatibility with clean room implementations.
ARM has patents on various parts of the ISA itself (the same applies to x86: the main difference is x86 patents are well enough split between AMD and Intel they are essentially forced to cross-license). You can't make a compatible version without a patent license at least. (And the ISAs have evolved since they were first made, with new patents involved: you can now make a very old version of an ARM core without a license, but such a chip would be irrelevant for most anything today).
Afaik the Ex Apple folks did not use the ARM cores - they developed their own from scratch after the learnings they got from Samsung for their first ARM SoC in the iPhone and later their own A series.
> For years, ARM wanted to get its CPU instruction set into notebooks and desktop PCs. Now, for the first time, there is a realistic opportunity and ARM is torpedoing it to get a bigger slice of the pie.
What’s with the heavy handed pro-Qualcomm/anti-ARM bias in this news piece?
My interpretation is obvious breach of contract by Qualcomm. Seems very cut and dry from the facts laid out in this very same article.
I have no skin in this race, just thought it was odd.
> Seems very cut and dry from the facts laid out in this very same article.
It's not quite that simple.
Qualcomm has a separate license from ARM that allows doing all the things they are currently doing with Nuvia IP. ARM claims that this license does not cover Nuvia IP, and that since Nuvia broke the terms of their license, all that IP needs to be destroyed. Qualcomm claims that they bought the IP separately from any license, and that they can just continue using it under the terms of their own license, and that no term in that license prevents Qualcomm from using the Nuvia IP under it.
The issue is at least murky enough that no court has granted ARM an injunction over it.
> My interpretation is obvious breach of contract by Qualcomm
I’d be inclined to say the opposite based on the German article linked in this articles. Qualcomm has its own architecture license in addition to the now expired Nuvia one and can’t really figure out what arm is objecting to precisely aside from wanting money
Both sides are playing hardball. The likely outcome is probably a settlement and all the legal wrestling is just negotiation tactics. We're talking about two companies whose primary business model is licensing intellectual property; so they both know how to play this game very well. Neither Qualcomm nor ARM directly manufacturers or sells products; they both make money from licensing stuff to third parties. And it looks like Qualcomm is cutting some corners there and is now putting their customers on the spot with this. Which no doubt is not going down well with those.
So, my guess is that Qualcomm will try to eventually fix this by throwing money at it and striking some deal with ARM. The question is just going to be how much. ARM is just twisting their arm (no pun intended) with this. But I agree that at face value they are well within their rights to escalate this. And given that their core business is selling ARM licenses, making sure that they get their money from Qualcomm kind of is their core business.
> What’s with the heavy handed pro-Qualcomm/anti-ARM bias in this news piece?
Pro shiny new devices and moving Windows into an ARM future, vs. Anti throwing away loads of work/tech over IP squabbles/greed (whoever is responsible for it)?
From the environmental point of view, it's incredible that they want these machines, built with so much energy and resources (e.g. a chip fab uses a lot of water [1], and a laptop is more than just chips), to just be destroyed...
Not to mention the planned obsolecence of non-upgradeable machines with near-unreplaceable battery packs (a very popular 'dark pattern' these days in device design, clearly aimed at encouraging people to replace the entire device more frequently)
The new ARM laptops are looking very promising: similar to the new ARM macbooks. So PC tech enthusiasts/media are gonna be on Qualcomm's side here, at least while it's framed as 'ARM is trying to stop these from actually being sold'.
The industry is looking and taking notes: Qualcomm a major IP holder doesn't want to pay for IP while ARM prefer money over the success of its ecosystem.
RISC-V immediately comes to mind as a long term alternative. Thank you ARM and Qualcomm.
Chromebooks are not GNU/Linux, rather a browser running on top of a POSIX kernel that happens to be Linux kernel, and could be Zirkon or a BSD tomorrow if Google felt so inclined, with the ability to run GNU/Linux workloads inside of a VM.
It is as desktop as running a Linux distribution inside Virtual Box.
> Chromebooks are not GNU/Linux, rather a browser running on top of a POSIX kernel that happens to be Linux kernel
That happens to be build using gcc, glibc and wayland. If you enable "developer mode" you get access to a subset of coreutils like "cp". You also get "chromebrew" which allows you to install the rest.
I think it's fair to say it's GNU/Linux stripped back to the bare essentials, with a custom desktop and a few GUI apps added. And the ability to add a fair chunk of the stripped stuff back. https://www.chromium.org/chromium-os/developer-library/guide...
> could be Zirkon or a BSD tomorrow if Google felt so inclined, with the ability to run GNU/Linux workloads inside of a VM.
Yes, well Windows could also be a Zirkon or a BSD tomorrow, with the ability to run Windows workloads inside of a VM.
Would it be any better if Qualcomm developed their own proprietary RISC-V cores with proprietary/patented extensions that most software would be reliant on? Currently Qualcomm and ARM are somewhat balancing each other out, with RISC-V if Qualcomm got significantly ahead they'd probably be in a position where it would be impossible for anyone to compete against them (I can't really see any reason for them to make their extensions freely available to their competitors).
They helpfully send the whole article before asking you to enable tracking cookies. You can e.g. use reader mode to read the article. This mostly seems like an oversight on their part though.
As mentioned in the article the site maliciously complied by (barely) changing the behaviour, which meant that the complaint has been renewed:
> Felix Mikolasch, Data Protection Lawyer at noyb: “It seems that the LfD's decision has already been overtaken by the latest version of the "Pay or Okay" banner. We will of course go back to the LfD and continue to fight this practice.”
It is not. By subscribing you uniquely identify yourself, with your real-world identity. Is is not necessary to use tracking cookies, when you log-in with your credentials.
Yep, you can only select "Zustimmen (agree)" for "Datenverarbeitungen von Werbeanbietern einschl. personalisierter Werbung mit Profilbildung (data processing by advertising providers including personalized advertising with profiling)."
I was disappointed to find this out after navigating through text written in an unfamiliar language and a complicated cookie dialog comparable to this [1]. German cookie dialog on an English article, is that even legal?
Without going into the details, I suspect this is all Qualcomm's fault. There's a reason that Apple had to go in-house to get a high quality ARM system working.
My guess is that, when Nuvia was bought off by Qualcomm, the license from ARM was no longer valid, since Nuvia ceased to exist, or rather, the license is non-transferable.
The conflict is over whether the license is transferable. Arm never would have sold that license to Qualcomm, especially with those terms, especially for that purpose.
Qualcomm bought a company because they had a license arm wouldn’t sell them. It comes down to subtleties in the contracts.
No, Qualcomm had already acquired an arch license from ARM, they had one before the purchase. The issue is that the Nuvia arch license had extra terms that the Qualcomm one didn't. ARM claims that the IP created under the Nuvia license cannot be transferred to be under the Qualcomm license. Qualcomm contends that it can.
> No, Qualcomm had already acquired an arch license from ARM, they had one before the purchase
I know they were making their own cores several years ago but for a bunch of design cycles they used ARM reference cores. during that time they were still an architectural licensee? Or they let it lapse and then re-licensed just before/after the Nuvia acquisition?
Qualcomm bought Nuvia not because they wanted the license Nuvia had, because that license was server-only, with a "don't compete with our stuff on mobile" carveout, but because they wanted the IP. The license that Qualcomm has is strictly better than what Nuvia had, and no-one wants the Nuvia license.
>How it happened and there was not a clause in the licensing agreement to explicitly cover these cases?
There were license clauses, the two companies disagree about what they meant, they're currently suing each other over them. This article is about that lawsuit and recent developments.
The issue is the application limitation of the license.
ARM granted Nuvia the license for server applications. Typically that license transfers, but ARM is taking umbrage with Qualcomm using the Nuvia developed tech (licensed for server applications) and applying it to consumer / handheld.
Did you read the article? It seems like ARM is against Qualcomm using a custom ARM core developed by Nuvia for licensing reasons.
Clearly the "off the shelf" processor designs ARM is producing aren't getting anywhere near the performance that Apple is getting from their own designs. Qualcomm are trying to compete by using the Nuvia design, and ARM aren't happy.
I have to wonder what value ARM are actually adding here. They want more money (doesn't everyone), but if they're unable to actually design performant processors then all they're really licensing is the instruction set. How long until the manufacturers realise that they can switch to RiscV and not pay ARM anything?
I don't think RISC-V is mature enough to warrant manufacturers moving to this ISA. If this was the case, we'd already have powerful RISC-V portable computers.
Qualcomm certainly have some interest in RISC-V, in fact there's a bit of a rift in the ecosystem in part because they are on one side in terms of how the design should go. But that means it's still very early days in terms of anything at the complexity/performance level of these cores being manufactured.
SoftBank is nickel and diming their customers to make up for the ARM sale to nVidia that fell through.
I think ARM on Windows will die eventually - what does it bring to the party ???.
Looking at Lunar Lake which is very much like the M series there is still much life left in that old dog :).
I suspect the short answer is "no", and the long answer is "if battery life is good enough who will care?"
My amd laptop from 3 years ago varies from 5-10 hours depending on workload. Compared to my mac its usually half the battery life.
But how many people need more than 5? They exist. But at that point the factors of price, performance, and compatibility are very relevant.
Reminds me of the state of phones now. Early smartphones had pathetic battery life. Now it's reached the point I don't care because the battery life of all the phones exceeds the daily use. Other factors now dominate.
... and this kind of nonsense is why x86 is alive, well, and likely to dominate the desktop space for decades.
Intel and AMD managed to sort out that bullshit decades ago, Intel even copied AMD's x64 extensions as Itanium fell flat.
It's not like ARM has something competitive with Intel in performance on their own - everyone but Apple and Annapurna/Amazon uses stock cores and they frankly suck. And instead of entering negotiations to copy from what either Apple or Amazon did with ARM to bring the entire ecosystem forward, they attempt to shoot Qualcomm. (The fact that Qualcomm itself sucks as a company makes all of that even more absurd)
The Intel fanboi (and thus x86 fanboi) in me is pleasantly surprised at this comedy show, but objectively this feels like something I should feel annoyed/saddened by.
I'm actually more of an ARM person, simply because neither Intel nor AMD managed to get power efficiency under control, not even after Apple's M1 ripped the x86 world to shreds.
The problem is, the ARM world never got away from crappy BSPs or managed to get any kind of standardization. Anyone reasonably skilled can get at least a basic "hello world" running that uses BIOS routines to draw characters on a screen and it will work on any x86 compliant machine, no matter its age - there is no equivalent that will work on ARM devices. None at all. Every tiny device, from embedded gadgets to full blown phones, has to have its own kernel build. What a mess that is.
You don't actually need a custom kernel build for every device these days (and haven't for years). Within the same ISA (ARM32 or ARM64) you can run the exact same binary kernel on multiple devices provided you have a suitable device tree for each.
And that's not really much different in principle to APCI (which is really "DT on steroids"). The main difference is that with ACPI the device manufacturer ships the tables whereas with DT they have usually have to be provided by others.
PCs haven't standardised everything at the hardware level for years. The original PC had standard IO addresses but this only covers a few things like keyboard, display, serial ports. Modern PCs, especially laptops, have all sorts of "non standard" chips hanging off I2C busses.
There are 2 reasons it "just works" on PCs
1) ACPI tables descibe it (just like DT)
2) There are compatibility tests needed to get the Microsoft hardware certification to run Windows
But it goes beyond the kernel to the full OS distribution. All PCs have pretty much the same use cases and similar UI capabilities. Sure some may be more powerful etc but they're all pretty much the same idea. So on a PC it makes snese to pop in a USB key with a "generic" Linux distribution an install it. It probably doesn't make as much sense to want to install the exact same distribution on your PC, phone, TV and router. This means that, given the OS is going to be customised anyway there just isn't as much demand for a fully plug and play system.
I think it's just never been a concern for embedded systems, which ARM has been used for mostly. Surely there is some standardisation coming, it'll just have to take a lot of time since there a lot of parties who need to come together.
But who actually uses that? That's my point. Phones all run their custom bootloaders (if you're lucky they use something from the Android SDK as second stage), embedded stuff runs u-boot versions forked a decade ago (or longer), even the really popular stuff such as raspi and its various clones all run their own proprietary stuff.
ACPI / UEFI is heavily used on ARM server systems.
Embedded stuff mostly uses u-boot true. But u-boot itself is actually fine for this type of device. I recently did an evaluation of several system on modules. They all came with their own usually different version of u-boot. For the purspose of the evaluation I ddin't care I just pointed them at my kernel + DT and it booted.
The "decades old" part really depends. At work I've just finished updating our u-boot 2020.10 to 2024.04.
But embedded and PC / Server are fundementally different in terms of use case. On a PC / Server the user expects to install an OS of his choice. Embedded isn't like that, the definition of embedded it pretty much "single use case device" so being able to install a generic Linux distro and run libre office isn't a use case many care about, particularly if it would increase the hardware costs.
Phones are a different case to embedded, here it *would* be useful to install a generic OS (though probably not the same distributions as on PCs). For that a standarised bootloader / OS interface would be useful but it deosn't have to be UEFI / ACPI, u-boot distroboot would be sufficient https://docs.u-boot.org/en/latest/develop/bootstd.html
I suspect only the apple laptops.
It was a pain for me to find an arm64 build of linux that'd run in normal vm software for example. Most were compiled for particular SBCs. Pretty much was left with only debian & opensuse.
So I agree with this topic: arm ecosystem has to get its act together to further adoption.
When Apple announced the M1 Pro/Max, I predicted that they would now hold a monopoly on good laptops for the next 5 years, based on where the PC market was at the time. (Good defined as high-performance, long battery life, and without other serious compromises such as weight)
We haven't seen independent reviews of these Qualcomm chips yet, but not even 3 years later and it sure looks like the entire PC market is about to leapfrog Apple. Intel's Lunar Lake looks like it's getting pretty close too.
Being a fanboi for a particular company/tech helps no-one except shareholders. Capitalists don't need to invest in competing if there's no-one competing with them.
I suspect this is more of an ambit claim from ARM to extract better terms from Qualcomm
> We haven't seen independent reviews of these Qualcomm chips yet, but not even 3 years later and it sure looks like the entire PC market is about to leapfrog Apple.
How? There is barely any software for Windows that is made for ARM, much less drivers for all the various kinds of gadgets people have or the various anticheat crapware, and without a solid runtime translation layer similar to Rosetta and backing for it in hardware (memory access orders, other acceleration stuff) any Windows effort will fall flat on its face as it did some years ago.
Qualcomm has shown off demos to the media of Baldur's Gate 3, video editing, music production, OBS, etc. Microsoft has claimed their translation layer is "as good as" Rosetta. Support for anti-cheat is not relevant unless you want to game online, which I don't. Driver support for third-party peripherals is a valid concern for a significant number of users, but not something I personally care about in a portable device.
It's perfectly reasonable to be sceptical of Qualcomm's claims, but you can get a feel for how confident a company is in their product by how they approach marketing - and we are seeing a level of confidence here from Qualcomm and Microsoft that we have not seen with previous Windows on ARM pushes.
I am personally inferring from this that the products are probably pretty good, but this is speculation. I won't be ordering one until I see some test results from independent reviewers.
Gamers are a major force behind technological advancements. AAA game makers didn't take mobile or Linux gaming seriously first until the Switch came along (that had enough firepower to run older AAA games) and finally Steam Deck provided actual console power in a mobile form factor.
Gamers are what gave Steam the financial incentive to get Proton to an entirely new level. And a platform that brands itself as the future must have gaming support.
Now of course, ARM is not happy about that. They spent resources to help Nuvia develop the core for servers. Now Qualcomm is using the Nuvia core to compete with stock ARM Cortex designs in phones, laptops, and servers.
ARM makes money from ISA and stock-ARM core licensing. They make way more money if you use their stock cores than if you only use their ISA. Oryon would only be using the ISA. Apple SoCs also use only the ISA.
Not only does ARM lose money from Qualcomm switching from Cortex cores to Oryon cores, they now have a legitimate competitor for Cortex SoCs other than Apple.
Qualcomm was also the the strongest lobbyist against the Nvidia/ARM deal. They don't want the competition from stock ARM cores backed by Nvidia. They don't want to compete with a combined Nvidia+ARM. They just want to compete with stock ARM cores with their Nuvia cores.
Other ARM vendors like Mediatek, Samsung have no ability to create better cores than Nuvia's cores. They can only rely on stock ARM core designs. Therefore, Qualcomm's Nuvia-derived cores should always be the high end* while other ARM vendors can only fight in the mid-range, which have much lower margins. Had Nvidia acquired ARM, Qualcomm would have been competing against ARM + Nvidia IP + Nvidia money.
I actually bought some Qualcomm stocks because I saw how Qualcomm outmaneuvered ARM.
*Assuming that stock ARM cores won't surpass Qualcomm's Nuvia-derived cores