News Qualcomm CEO expected to save $1.4 billion in Arm royalties by purchasing Nuvia, assumed Snapdragon X chips would be a massive hit

[Admin]

Qualcomm CEO Christiano Amon said that the company could save up to $1.4 billion in royalties to Arm by purchasing Nuvia.

Qualcomm CEO expected to save $1.4 billion in Arm royalties by purchasing Nuvia, assumed Snapdragon X chips would be a massive hit : Read more

 

[A Stoner]

The new processors are significantly better than any ARM processors that Qualcomm has access to, and the answer is not that ARM wants to get paid the royalties it thinks it is owed, but instead wants the technology destroyed. That is the story that needs to be taken away from this lawsuit, and where the judge and or jury should be looking at to see which side to support in their findings.

These designs do not infringe on ARM patents. Their argument is that they should get higher royalties for them. But their solution is to destroy technology instead.

 

[bit_user]

If Qualcomm could save at least $50 million annually because of its Nuvia purchase, other Arm clients could potentially copy this move

If it were easy to make competitive cores, sure.

This isn't the first time Qualcomm designed its own cores. As recently as the Snapdragon 820, they had their own fully-custom cores and they weren't bad. This is one of the reasons Qualcomm had an architecture license, BTW. However, they struggled to keep up with ARM's cores and so Qualcomm switched over to merely tweaking ARM's designs, starting with the Snapdragon 830.

Nuvia was sort of a unicorn, because you had an experienced CPU design team that was intimately familiar with the architecture and knew exactly how to build competitive mobile cores. That sort of thing only happens like once per decade, with the last time probably being when Apple bought PA Semi (which was building PowerPC cores, not ARM cores, but still aimed at laptops).

Samsung also used to have its own custom ARM cores. But, like Qualcomm, they struggled to compete with ARM's. So, about 5 years ago, Samsung killed off that group and switched to simply licensing ARM's IP.

 

[cyrusfox]

tricky... from what I can tell and what I understand, does appear Qualcomm is being deceptive and attempting to avoid royalties that on the surface are due. Perhaps what they are doing is legal, but I would not call it ethical.

The part about destroying technology, well if it based on ARM IP, you can understand where ARM is coming form, good luck to the courts, no confidence they get it right though. Maybe it would have been best for all if Nvidia was allowed to acquire ARM...

 

[thisisaname]

Only thing for certain is there will be an appeal.

 

[ezst036]

The new processors are significantly better than any ARM processors that Qualcomm has access to, and the answer is not that ARM wants to get paid the royalties it thinks it is owed, but instead wants the technology destroyed. That is the story that needs to be taken away from this lawsuit, and where the judge and or jury should be looking at to see which side to support in their findings.

These designs do not infringe on ARM patents. Their argument is that they should get higher royalties for them. But their solution is to destroy technology instead.

This is what makes the whole proposal so absurd though. The actual destruction of the blueprints/design. Even if Qualcomm has to fully end production/use of ARM altogether, that technology is very valuable. Qualcomm will never destroy it. Just sell it to someone else.

A part of me thinks this is ARM Holdings' way of shifting the Overton Window. They never actually want Qualcomm to destroy the design this whole time, but by making such a radical request,

"Oh hey, why don't you just make this reasonable request instead and give us $1.4 billion dollars, which is what the licensing fees would amount to anyways. Isn't that so much better?"

They are attempting to make their real goal this whole time seem much more reasonable.

What ARM Holdings doesn't seem to realize though is that by making the demand for design destruction, they as licensor are doing significant damage to their own brand. Who would want to license from this organization that either makes these sorts of demands or plays these sorts of gimmicky games when RISC-V is freely available?

If you're trying to make inroads on phones, ok, you gotta go ARM for now. That's how the entire software stack is built. But on desktops, laptops, or servers, it's wide open as to what technology will come out ahead of x86. It's not yet a lock that ARM will be the new standard.

 

[KraakBal]

To me it looks like ARM is butthurt about not having any licenceable designs to compete with and emulate x86.

They should have been working on these instead of sitting on their laurels in the 2010s

 

[bit_user]

Maybe it would have been best for all if Nvidia was allowed to acquire ARM...

Why do you think that would've changed anything, here?

 

[bit_user]

What ARM Holdings doesn't seem to realize though is that by making the demand for design destruction, they as licensor are doing significant damage to their own brand. Who would want to license from this organization that either makes these sorts of demands or plays these sorts of gimmicky games when RISC-V is freely available?

Oh, I'm sure they know. It could even be RISC-V that's motivating their behavior. They might see the days of AArch64 as limited either way, and are simply pursuing a revenue-maximizing strategy that more than makes up for shortening the transition point by disproportionately increasing revenues until then. It is risky, though.

Also, the only ones who really have to worry are those with architectural licenses. Most ARM customers are using ARM's cores, and thus not affected by disputes involving architecture licenses. The main exception to that is Apple, but I think I heard their architecture license is locked-in until like 2040. AMD and Nvidia reportedly also have architecture licenses, but I'm not sure when they're up. I think I heard something about Nvidia's getting extended before it put in its offer to acquire ARM.

If you're trying to make inroads on phones, ok, you gotta go ARM for now. That's how the entire software stack is built.

I think Android also supports RISC-V, now. There are some 3rd party native apps that would have to get ported, but most apps are JIT-compiled and run on all Android platforms.

But on desktops, laptops, or servers, it's wide open as to what technology will come out ahead of x86. It's not yet a lock that ARM will be the new standard.

Eh, servers have really gone pretty decidedly for ARM. RISC-V could ultimately win out, but not before ARM has even more growth in that market.

 

[Justin Goldberg]

Reminds me Transmeta IP being destroyed by Intel.

 

[dCasualGamer]

Nuvia was sort of a unicorn, because you had an experienced CPU design team that was intimately familiar with the architecture and knew exactly how to build competitive mobile cores. That sort of thing only happens like once per decade, with the last time probably being when Apple bought PA Semi (which was building PowerPC cores, not ARM cores, but still aimed at laptops).

Basically ARM just want to protect its own business model, its cash cow of licensing ARM instruction set. Whether the new chip design is better or competitive is beside the point. If Qualcomm is allowed to do this maneuver, then many in the industry will think of ways to do the same. Bunch of lawyers with fancy suits will find new ways to circumvent ARM's licensing contracts.

This is what makes the whole proposal so absurd though. The actual destruction of the blueprints/design. Even if Qualcomm has to fully end production/use of ARM altogether, that technology is very valuable. Qualcomm will never destroy it. Just sell it to someone else.

I guess the punishment has to be severe enough to deter future breach of contract. What is a million dollar fines if the resulting benefit is in the billions

 

[thestryker]

However, they struggled to keep up with ARM's cores and so Qualcomm switched over to merely tweaking ARM's designs, starting with the Snapdragon 830.

I think there's certainly an argument to be made about shortsighted management in making this move. Qualcomm's design arm seemed to be pretty good and I think given the opportunity could have recovered on their own. Of course at that time Qualcomm wasn't commanding the level of price premium they are now.

Samsung on the other hand really seemed to drop the ball on multiple generations and had some very serious problems with their designs.

 

[stuff and nonesense]

We aren’t privy to the licence terms granted to Nuvia and Qualcomm.

However

Id think it would be difficult for Qualcomm to sell the parts if the licence were to be revoked as they would at that point be trying to make money from a product to which they would have had their access removed.

If the idea behind buying Nuvia (apart from acquiring a technology/design) was to avoid/evade royalty payments I’d argue that ARM is well within their rights and should win the case.
Their business model, not a cash cow, is to design and license. It has worked well over a number of of years with few if any (publicised) court cases.

Manufacturers and ARM have cooperated in a mutually beneficial arrangement. Qualcomm seemingly trying to avoid a potential $1.4 billion suggests a strong motive to break the arrangements.

[Edit]

Qualcomm claim in the article that there is a negligible amount of ARM tech in the Nuvia derived CPUs - I see the Qualcomm claim as a red herring as ARM licenses the instruction set (and assists with chip design).

Consider- how much AMD tech is in the AMD64 compliant Intel Core 2 and later CPUs?
Intel and AMD cross license access to x86, x86-64 and its extensions but it is up the “competitor” to implement the instructions. The instruction set is the valuable ip.

 

[thestryker]

If the idea behind buying Nuvia (apart from acquiring a technology/design) was to avoid/evade royalty payments I’d argue that ARM is well within their rights and should win the case.

It's not made clear in this article, but the royalty difference in question is related to the selling products under a technology license agreement versus an architecture license agreement. Qualcomm has been using modified Arm cores under a TLA for around 8 years now which carries a higher royalty rate than custom cores under an ALA.

 

[stuff and nonesense]

It's not made clear in this article, but the royalty difference in question is related to the selling products under a technology license agreement versus an architecture license agreement. Qualcomm has been using modified Arm cores under a TLA for around 8 years now which carries a higher royalty rate than custom cores under an ALA.

Thanks for clarifying.

 

[evdjj3j]

Qualcomm CEO expected to save $1.4 billion in Arm royalties by purchasing Nuvia, assumed Snapdragon X chips would be a massive hit​

Really? He thought that?

 

[Steve Nord_]

Qualcomm CEO Christiano Amon said that the company could save up to $1.4 billion in royalties to Arm by purchasing Nuvia.

Meanwhile there are... 0 loud twisty suits about Risc V Bus or DDR5 interfaces or instruction set extensions, M core extensions for embedded, or compute-in-DDR5? Is it Mentor user forum stuff anymore?

 

[bit_user]

I think there's certainly an argument to be made about shortsighted management in making this move. Qualcomm's design arm seemed to be pretty good and I think given the opportunity could have recovered on their own. Of course at that time Qualcomm wasn't commanding the level of price premium they are now.

Chips and Cheese did a deep dive to tease out its strengths and weaknesses, not long ago:

https://chipsandcheese.com/p/kryo-qualcomms-last-in-house-mobile-core
​

For the TL;DR, just read their conclusion:

https://chipsandcheese.com/i/138977268/final-words
​

It does seem to me like they could've continued to at least keep up with ARM, but the weaknesses highlighted by the article suggest that maybe they weren't entirely up to the task they set before themselves. Perhaps devoting so many resources to the CPU cores also had opportunity costs, elsewhere. If switching to ARM's cores enabled them to focus more resources on the rest of the SoC, who am I to say it was a bad decision?

That said, if their activist investors weren't so keen on getting a bigger dividend, maybe Qualcomm could've continued to invest in all aspects of their SoCs and wouldn't even have been in a position of needing Nuvia.

 

[thestryker]

Chips and Cheese did a deep dive to tease out its strengths and weaknesses, not long ago:

​

https://chipsandcheese.com/p/kryo-qualcomms-last-in-house-mobile-core​

Yeah I read that back when they published it and it's enlightening.

It does seem to me like they could've continued to at least keep up with ARM, but the weaknesses highlighted by the article suggest that maybe they weren't entirely up to the task they set before themselves. Perhaps devoting so many resources to the CPU cores also had opportunity costs, elsewhere.

Keep in mind that was also developed at the same time as Centriq which used a different design. I can't help but wonder if that decision ended up hurting CPU design more than expected or if Arm's designs ended up better than expected.

From a business standpoint that was also right in the middle of all the NXP and Broadcom acquisition issues.

 

[bit_user]

Meanwhile there are... 0 loud twisty suits about Risc V Bus or DDR5 interfaces or instruction set extensions, M core extensions for embedded, or compute-in-DDR5?

Does RISC-V specify a bus standard? I thought it was just ISA and software-visible system architecture stuff.

Sure, some of the ISA extensions could be subject to litigation, if they're not adequately protected by patents in the pool (I'm not sure how RISC-V usually does it, but standards normally work by having member companies agree to pool their patents and agree not to sue other implementers for infrigement).

Who would be suing over DDR5 interfaces, Rambus? DDR5 isn't even that new, any more, so you'd think anyone with infringement claims would've stepped forward, by now.

Compute-in-memory is still yet to take hold, so that is indeed an area where there could be a snake in the grass, waiting for a ripe time to strike.

 

[bit_user]

Keep in mind that was also developed at the same time as Centriq which used a different design. I can't help but wonder if that decision ended up hurting CPU design more than expected or if Arm's designs ended up better than expected.

I think they were more serial than concurrent. Snapdragon 820 shipped in December 2015. Centriq started sampling in late 2018. Perhaps Centriq did play into Qualcomm's decision to end mobile core development, but then the just pivoted to focusing on Centriq?

 

[Mama Changa]

To me it looks like ARM is butthurt about not having any licenceable designs to compete with and emulate x86.

They should have been working on these instead of sitting on their laurels in the 2010s

I'll bet a bottom dollar the rumours about ARM wanting to get into chip production, that ARM are doing everything in their power to dis_ARM the strongest player in this market so they can have as little competition as possible. It was shameless lie from ARM the other day saying they have no interest in building their own chips. Everybody in the industry say it's coming.

 

[thestryker]

I think they were more serial than concurrent. Snapdragon 820 shipped in December 2015. Centriq started sampling in late 2018. Perhaps Centriq did play into Qualcomm's decision to end mobile core development, but then the just pivoted to focusing on Centriq?

It was announced in late 2014 and first demonstrated in 2016 so it's hard to tell whether there would be overlap, a pivot, or both.

 

[bit_user]

It was shameless lie from ARM the other day saying they have no interest in building their own chips. Everybody in the industry say it's coming.

It was sometime last year that I think ARM basically announced they wanted to start making/selling CPU tiles, with the idea that their customers could integrate ARM cores via chiplets, instead of having to deal with fabbing that part of the SoC themselves.

 

[Amdlova]

And people say intel is evil
Qualcomm loose the chance of get the mostly evil company known by man.
Now back to scratch...