Solved! Qualcomm buys Nuvia for $1.4 Billion

[Gideon]

Quite an absurd valuation for a startup without a single project yet.

AnandTech Forums: Technology, Hardware, Software, and Deals

Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

www.anandtech.com

Anyway, on top of Server chips Qualcomm should again have top-performing custom ARM cores for mobile SoCs in the future (judging by the development team, possibly better than Apple)

EDIT: and looks like laptops as well

NUVIA CPUs are expected to be integrated across Qualcomm Technologies' broad portfolio of products, powering flagship smartphones, next-generation laptops, and digital cockpits, as well as Advanced Driver Assistance Systems, extended reality and infrastructure networking solutions.

 

[DrMrLordX]

Or Nuvia rapidly found out that the ARM server market just isn't suitable for merchant silicon vendors, and none of the big cloud providers wanted to purchase them.

Ampere is still trying to make it in that space. Maybe they're waiting for a buyout too?

At least now we know that Nuvia's claims are likely legit. And it would not surprise me to see Nuvia-inspired Qualcomm chips in Microsoft's future Surface products.

 

[NTMBK]

AnandTech Forums: Technology, Hardware, Software, and Deals

Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

www.anandtech.com

It's a done deal, and Qualcomm are aiming to have laptop chips with Nuvia cores out by the end of next year. Looking forward to this!

 

[Doug S]

Interesting that they are doing laptop first, with no mention at all of phones and a bit of lip service to servers but no firm plans to re-enter that market. Qualcomm must really believe there's a market for high end ARM laptops. The market for a high end Chrome/Linux laptop would be extremely limited, they must be hoping they can "make fetch happen" and Windows on ARM will become something people actually care about. Good news if Apple if it does, bad news for Intel.

Doesn't bode well for Nuvia's core team sticking around once their lockup period is expired if Qualcomm only lets them "explore" servers but chooses not to re-enter that market, given that that was supposedly why they left Apple. I guess with all the money they made from selling Nuvia, if they do another startup they'll be able to do it on their own dime without VCs this time around.

 

[NTMBK]

Interesting that they are doing laptop first, with no mention at all of phones and a bit of lip service to servers but no firm plans to re-enter that market. Qualcomm must really believe there's a market for high end ARM laptops. The market for a high end Chrome/Linux laptop would be extremely limited, they must be hoping they can "make fetch happen" and Windows on ARM will become something people actually care about. Good news if Apple if it does, bad news for Intel.

If they can get Apple M1 style performance, I think they can make it happen. If it runs emulated x86 apps as fast (or faster) than an Intel chip, at lower power, with even higher performance on native code? That's a pretty compelling CPU.

Doesn't bode well for Nuvia's core team sticking around once their lockup period is expired if Qualcomm only lets them "explore" servers but chooses not to re-enter that market, given that that was supposedly why they left Apple. I guess with all the money they made from selling Nuvia, if they do another startup they'll be able to do it on their own dime without VCs this time around.

Depends on whether the core team have the appetite to fail again. The market for merchant ARM server CPUs just doesn't seem to be there- how many companies have thrown in the towel at this point? The real money is in the hyperscalers, and they want to design their own chips.

 

[Doug S]

If they can get Apple M1 style performance, I think they can make it happen. If it runs emulated x86 apps as fast (or faster) than an Intel chip, at lower power, with even higher performance on native code? That's a pretty compelling CPU.

There's little chance they can do static translation like Rosetta 2 does. Apple had years to prepare for it and a developer monoculture which means almost every Mac x86 app was compiled using Apple's tools - and they had successfully dropped 32 bit x86 prior to the introduction of the M1 Macs.

Basically Apple had every advantage, and probably all were needed to make this happen. These Qualcomm ARM laptops will have to use Microsoft's JIT translator which are much slower than native speed.

 

[Doug S]

Depends on whether the core team have the appetite to fail again. The market for merchant ARM server CPUs just doesn't seem to be there- how many companies have thrown in the towel at this point? The real money is in the hyperscalers, and they want to design their own chips.

How can they "fail again" when they haven't failed a first time? They never got a chance to bring their chips to market, so who knows how well they would do. The hyperscalers will buy from someone else if the performance/watt advantage was sizeable enough.

 

[NTMBK]

How can they "fail again" when they haven't failed a first time? They never got a chance to bring their chips to market, so who knows how well they would do. The hyperscalers will buy from someone else if the performance/watt advantage was sizeable enough.

It's not like Qualcomm has launched a hostile takeover. They chose to sell to a company that has no interest in ARM servers. If the demand was there for merchant ARM server SoCs, they would not have sold to Qualcomm.

They set out to prove that they could make it as an ARM server vendor, and in that goal they have failed.

 

[Thala]

There's little chance they can do static translation like Rosetta 2 does. Apple had years to prepare for it and a developer monoculture which means almost every Mac x86 app was compiled using Apple's tools - and they had successfully dropped 32 bit x86 prior to the introduction of the M1 Macs.

Basically Apple had every advantage, and probably all were needed to make this happen. These Qualcomm ARM laptops will have to use Microsoft's JIT translator which are much slower than native speed.

Microsofts JIT is between 50% and 70% native speed, while Apples static translation is between 60% and 85%. I would not qualify this as much slower. In any case emulation is just a stop-gap for those apps, which are not available as ARM64 and of course for legacy apps.

 

[scannall]

It's not like Qualcomm has launched a hostile takeover. They chose to sell to a company that has no interest in ARM servers. If the demand was there for merchant ARM server SoCs, they would not have sold to Qualcomm.

They set out to prove that they could make it as an ARM server vendor, and in that goal they have failed.

I'm of the opinion Nuvia was formed to be bought out. Make noise, develop some IP and get a big payday. Their business model was rudimentary and unworkable any other way.

 

[Doug S]

Microsofts JIT is between 50% and 70% native speed, while Apples static translation is between 60% and 85%. I would not qualify this as much slower. In any case emulation is just a stop-gap for those apps, which are not available as ARM64 and of course for legacy apps.

Where do you see these 50-70% native for JIT? The only time that's true is for something that runs a ton of native library code. Typically people assume about 30-50% native for JIT, and the former is more likely than the latter.

Measurements I've seen from people running real world code on Windows ARM have been closer to the 30% mark, with only a few outliers cracking 50%.

 

[Thala]

Where do you see these 50-70% native for JIT? The only time that's true is for something that runs a ton of native library code. Typically people assume about 30-50% native for JIT, and the former is more likely than the latter.

Measurements I've seen from people running real world code on Windows ARM have been closer to the 30% mark, with only a few outliers cracking 50%.

I did run a few benchmarks and compared the numbers with native code. I did use benchmarks, which are limited by emulation performance like 7-zip for instance. It also reflects my observations when porting applications to Windows ARM64 from x64.

 

[soresu]

Typically people assume about 30-50% native for JIT, and the former is more likely than the latter.

Derp, less than 30-50% I should think.

Going by ASM.js being AOT and getting in that region of perf I should be surprised at even the most turbo charged optimising (IR utilising) JIT making 30+% of native performance.

People tend to assume that 'native code' automatically means ARM code equally optimised to the equivalent native x86 code - which given the greater age of x86 implementations seems pretty unlikely in many places.

I've seen it stated before that the ARMv8 dynarec backend for Dolphin is still behind the x64 backend, even after several years of dev time.

In the same fashion the GCC and LLVM compiler backends for C -> ARM native may not always be entirely equal in optimisations.

 

[NTMBK]

Latest update on this:

In statements to Reuters, Amon had made comments regarding the company’s future CPU roadmap, which come to further contextualise the company’s completed acquisition of NUVIA last March.

"We needed to have the leading performance for a battery-powered device," Amon said. "If Arm, which we've had a relationship with for years, eventually develops a CPU that's better than what we can build ourselves, then we always have the option to license from Arm."

AnandTech Forums: Technology, Hardware, Software, and Deals

Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

www.anandtech.com

Genuinely looking forward to seeing laptops with Nuvia cores in them.

 

[gdansk]

That's a very promising statement. He thinks Nuvia's design is better than ARM's latest and perhaps next generation too.

"Amon now stating that they are planning on bringing such a design to market in 2022."
That seems pretty soon. Wonder if that's code for a Q4 paper launch.

It's strange feeling being optimistic about Qualcomm.

 

[soresu]

He thinks Nuvia's design is better than ARM's latest and perhaps next generation too.

That "perhaps" part is the most interesting to me for the ARM community as a whole.

It implies that he thinks their first gen Phoenix core may be stretching it to even match Cortex X3.