News Arm Sues Qualcomm and Nuvia for Breaking License Agreement

[Admin]

Arm accuses Qualcomm and Nuvia of illegally developing Arm processors following Nuvia takeover.

Arm Sues Qualcomm and Nuvia for Breaking License Agreement : Read more

 

[suryasans]

Qualcomm should use x86-64 ISA for its upcoming SoCs since AMD64 patents have been expired. It's highly likely a very respected company like Qualcomm will get a comprehensive cross lisensing agreement from both Intel and AMD without paying expensive royalties like what has been happened to ARM architectural licensees.

 

[watzupken]

Looks like Qualcomm is walking on thin ice for their ambition to release their desktop class ARM SOC.

 

[christopher.andrew.carr]

Qualcomm should use x86-64 ISA for its upcoming SoCs since AMD64 patents have been expired. It's highly likely a very respected company like Qualcomm will get a comprehensive cross lisensing agreement from both Intel and AMD without paying expensive royalties like what has been happened to ARM architectural licensees.

In that case they would not be able to get the performance/power usage which is the entire purpose of their new SoC.

I imagine they will come to some arrangement with ARM.

High-performance RISC-V laptops will be nice, some day.

 

[TerryLaze]

Qualcomm should use x86-64 ISA for its upcoming SoCs since AMD64 patents have been expired. It's highly likely a very respected company like Qualcomm will get a comprehensive cross lisensing agreement from both Intel and AMD without paying expensive royalties like what has been happened to ARM architectural licensees.

LOL in what world do you live in?!
Intel is never going to give away any of their IP to anybody,
and AMD can't afford to have another competitor period.

Building a CPU based solely on expired patents is as lucrative as it sounds, you will be spending countless amounts of money to develop something that stopped being relevant decades ago.

 

[bit_user]

Wow, and I thought Apple would be the one to take them down.

Anyway, I highly doubt this will derail Qualcomm/Nuvia. They have too much at stake and ARM knows it. It's just an extortion play, and a dangerous one at that. ARM needs to pad its revenues in preparation for its upcoming IPO, but if they're too aggressive about this sort of thing, it could put off other licensees.

I don't have one iota of pity for Qualcomm. I haven't followed their various legal issues, over the years, but I gather they're the aggressor in more than 50% of the cases. I think the ones ultimately at fault are the Wall St. firms pushing Qualcomm to monetize every last cent out of their IP. That greed permeated its way into Qualcomm's corporate culture, causing them to underinvest in their own design teams, which is why they had to go out and buy someone like Nuvia to once again have a competitive core design.

 

[bit_user]

LOL in what world do you live in?!

You missed the most obvious problem with that idea, which is that Intel already tried using x86 for phones (Qualcomm's main business), and it failed spectacularly.

 

[renz496]

Qualcomm should use x86-64 ISA for its upcoming SoCs since AMD64 patents have been expired. It's highly likely a very respected company like Qualcomm will get a comprehensive cross lisensing agreement from both Intel and AMD without paying expensive royalties like what has been happened to ARM architectural licensees.

Lol if that's the case nvidia already have their x86 CPU right now. Intel sour the cross licensing deal they had with nvidia in 2004 out of the fear nvidia will make their own x86 CPU. On the surface it is about the chipset nvidia make for intel cpu but the settlement in 2011 confirms about intel fear about nvidia making x86 cpu. Also intel most definitely not forgetting the anti competitive thing Qualcomm did with their "Snapdragon package" on mobile market before which in the end billions that intel spend to enter that market just go up in the flames. Intel for their part has been doing everything in their power to not make another x86 player enter the market

 

[renz496]

You missed the most obvious problem with that idea, which is that Intel already tried using x86 for phones (Qualcomm's main business), and it failed spectacularly.

And intel failed on mobile (smartphones) mainly because of Qualcomm not so much because of ARM is better than intel x86.

 

[bit_user]

And intel failed on mobile (smartphones) mainly because of Qualcomm not so much because of ARM is better than intel x86.

Is that why Intel's SoCs had so few takers?

Look, it's no secret that x86 just doesn't scale down well to low power. Apart from the phone market, Intel also tried to tackle the IoT market with its Edison product line. That didn't last long, either.

 

[watzupken]

Wow, and I thought Apple would be the one to take them down.

Anyway, I highly doubt this will derail Qualcomm/Nuvia. They have too much at stake and ARM knows it. It's just an extortion play, and a dangerous one at that. ARM needs to pad its revenues in preparation for its upcoming IPO, but if they're too aggressive about this sort of thing, it could put off other licensees.

I don't have one iota of pity for Qualcomm. I haven't followed their various legal issues, over the years, but I gather they're the aggressor in more than 50% of the cases. I think the ones ultimately at fault are the Wall St. firms pushing Qualcomm to monetize every last cent out of their IP. That greed permeated its way into Qualcomm's corporate culture, causing them to underinvest in their own design teams, which is why they had to go out and buy someone like Nuvia to once again have a competitive core design.

Considering that Qualcomm have deliberately delay and buying time, instead of working something out with ARM, I am pretty sure they will play the same game in the courts. There is a lot of stake for Qualcomm to just destroy what the Nuvia team were working on. At the same time, I feel they are unsure of the success of the product, and so would make sense for them to drag it out to see if it is worth paying.

 

[renz496]

Is that why Intel's SoCs had so few takers?

Look, it's no secret that x86 just doesn't scale down well to low power. Apart from the phone market, Intel also tried to tackle the IoT market with its Edison product line. That didn't last long, either.

Not to the point it cannot be used in phones or tablet. I remember Asus ZenFone in the past use intel soc for quite sometime. Also in the past some ARM engineer or rep has done interviews here at toms. He was ask something about ARM efficiency over x86 and i still remember what the guy reply in general: while he happy that people think that ARM is more efficient than x86 in reality he said their isa/architecture is not more efficient than x86.

The main reason phone maker not picking intel soc is because of modem. Majority of phone maker are not really receptive towards software based modem that intel have vs the more traditional hardware based one. Nvidia saw this issue early on so they decided to ditch the effort on smartphone right just after acquiring Icera. Jensen back then said the only soc maker (that also not makes phone) that can really competing in this market is Qualcomm and Mediatek since both own hardware based modem. Things get more complicated when Qualcomm are forcing their snapdragon package if phone maker want soc with their baseband/modem. This move also what makes samsung end up feeling bitter towards qualcomm because they no longer able to license qualcomm modem for their exynos. Snapdragon or you can't sell your phones in a market where you need qualcomm baseband/modem. Apple is the only one qualcomm can't shove their snapdragon to so they decided to charge more expensive licensing price vs their other licensee.

 

[TerryLaze]

You missed the most obvious problem with that idea, which is that Intel already tried using x86 for phones (Qualcomm's main business), and it failed spectacularly.

That was quite a while ago though. When people were only running very basic stuff on their phones.
Now they are starting to expect to run windows and heavy duty emulation (switch) and whatnot on their phones and it's difficult to scale ARM up to run them fast enough.
x86 might still fail just as badly now (if anybody would try) ,but things do have changed more towards the favor of x86.

 

[bit_user]

There is a lot of stake for Qualcomm to just destroy what the Nuvia team were working on. At the same time, I feel they are unsure of the success of the product, and so would make sense for them to drag it out to see if it is worth paying.

In the technology business, you cannot afford to drag out product development. Time to market is nearly everything. If a chip launches too late, its tech will be obsolete and the market will already be ahead of it.

 

[bit_user]

in the past some ARM engineer or rep has done interviews here at toms. He was ask something about ARM efficiency over x86 and i still remember what the guy reply in general: while he happy that people think that ARM is more efficient than x86 in reality he said their isa/architecture is not more efficient than x86.

If you don't put a timestamp on that, it loses any hope of relevance.

What I'm talking about is data. Look at perf/W of Ampere Altra or Amazon Graviton 3 vs. AMD Milan. That's a heck of a lot more relevant than what some random person said, some unspecified number of years ago, about some unspecified version of ARM ISA and set of ARM cores.

 

[bit_user]

That was quite a while ago though. When people were only running very basic stuff on their phones.
Now they are starting to expect to run windows and heavy duty emulation (switch) and whatnot on their phones and it's difficult to scale ARM up to run them fast enough.
x86 might still fail just as badly now (if anybody would try) ,but things do have changed more towards the favor of x86.

ARM-based server CPUs suggest that ARM can scale up just fine.

In spite of using a lot less power and lacking SVE optimizations for several benchmarks, Graviton 3 holds its own against AMD and Intel:

Source: https://www.phoronix.com/review/graviton3-amd-intel

 

[TerryLaze]

ARM-based server CPUs suggest that ARM can scale up just fine.

Different type of scaling...
Getting 10 women pregnant at the same time isn't going to give you a baby instantly, each baby is still going to take 9 months to be produced.

The same goes for CPUs, sticking a bunch of ARM cores together doesn't make each core any faster and the things I talk about cant be broken up into infinite parts they are mainly single thread heavy, faster cores is what you need to run windows, emulators and a bunch of other stuff that people are starting to expect on smartphones, fast enough.

Making individual ARM cores as fast as individual x86 cores in the type of code we are talking about is extremely difficult.

 

[bit_user]

Different type of scaling...
Getting 10 women pregnant at the same time isn't going to give you a baby instantly, each baby is still going to take 9 months to be produced.

The same goes for CPUs, sticking a bunch of ARM cores together doesn't make each core any faster

It's not, though. 80-core/80-thread Ampere Altra holds its own against an Epyc Milan 64-core/128-thread CPU. That shows it's not only competing by virtue of sheer core count.

In the Amazon Graviton 3 benchmarks I cited, the size of the VMs was the same across all CPUs. Now, the catch is that a vCPU on Graviton 3 is a dedicated core, whereas it's just 1 hyperthread on AMD and Intel. However, the entire 64-core Graviton 3 is using only 100 W, which amounts to much less per vCPU than on AMD or Intel.

Also, there's what I mentioned about some of the benchmarks being optimized for AVX, but not ARM NEON/SVE. If you exclude those, the results would be very much in favor of Graviton 3.

Making individual ARM cores as fast as individual x86 cores in the type of code we are talking about is extremely difficult.

And yet, they seem to be getting there. Graviton 3 is derived from X1/A78 cores, which are now 1 generation old and 2 generations behind what's coming to devices within the next year. The Ampere Altra I mentioned above is a further couple generations behind, being based on N1 cores derived from Cortex-A76.

In terms of microarchitecture parameters, the X3 seems like it's really starting to close the gap with x86.

Arm Unveils Next-Gen Flagship Core: Cortex-X3

Arm announces its next-generation flagship core: the Cortex-X3.

fuse.wikichip.org

 

[TerryLaze]

It's not, though. 80-core/80-thread Ampere Altra holds its own against an Epyc Milan 64-core/128-thread CPU. That shows it's not only competing by virtue of sheer core count.

In the Amazon Graviton 3 benchmarks I cited, the size of the VMs was the same across all CPUs. Now, the catch is that a vCPU on Graviton 3 is a dedicated core, whereas it's just 1 hyperthread on AMD and Intel. However, the entire 64-core Graviton 3 is using only 100 W, which amounts to much less per vCPU than on AMD or Intel.

Yes, the fact that arm is better for "extremely little power per core" was never in question.

Give a single arm core access to 40W or whatever a modern x86 cpu uses for single thread and show me that it can compete at that.

 

[bit_user]

Give a single arm core access to 40W or whatever a modern x86 cpu uses for single thread and show me that it can compete at that.

It's true! I think nobody has yet built an ARM core that can burn the same amount of power as Alder Lake's Golden Cove!
; )

 

[TerryLaze]

It's true! I think nobody has yet built an ARM core that can burn the same amount of power as Alder Lake's Golden Cove!
; )

And that's why they will never be a danger for desktop CPUs.

 

[ikjadoon]

Yes, the fact that arm is better for "extremely little power per core" was never in question.

Give a single arm core access to 40W or whatever a modern x86 cpu uses for single thread and show me that it can compete at that.

Did I wake up in 2012? Do you keep up with any non-x86 CPUs?

The A15 P-core needs 4.11W to hit 7.28 SPECint2017 points, while the Ryzen 5950X needs 20W+ for 0.01 more. All of today's x86 cores are significantly narrower with much less IPC than any high-perf Arm core. Luckily, Arm cores now hit 3 GHz easily: massive IPC + solid clocks = an iPhone 13 can run 1T workloads faster than 99.9% of desktop PCs (assuming, at most, 0.1% of PCs are i7/i9 Alder Lakes).

This reads like, "Until your CPU needs an 360mm AIO, it can't be faster than my CPU!"

 

[bit_user]

And that's why they will never be a danger for desktop CPUs.

Never is a long time. I want to see what Apple does to replace their Mac Pro.

 

[bit_user]

...

Seeing that chart really makes me miss Andrei. We're going to be sorely lacking, with him gone from Anandtech and out of the journalism gig.

At least Ian is still publishing on Twitter and elsewhere, but I haven't tried to follow.

Did you notice Hot Chips 2022 came & went, with nary a mention on Anandtech?

 

[TerryLaze]

Did I wake up in 2012? Do you keep up with any non-x86 CPUs?

The A15 P-core needs 4.11W to hit 7.28 SPECint2017 points, while the Ryzen 5950X needs 20W+ for 0.01 more.

Yes, but if you would rip everything out of an AMD CPU that arm can't do either would that difference still be that big?
You can't say "the A15 is running that benchmark just as well as X only that it is running a completely different benchmark... "

One continuing issue with SPEC CPU 2017 is the Fortran subtests; due to a lacking compiler infrastructure both on iOS and Android, we’re skipping these components entirely for mobile devices. What this means also, is that the total aggregate scores presented here are not comparable to the full suite scores on other platforms, denoted by the (C/C++) subscript in the score descriptions.