I was really excited about Centriq - talked to some people who had spent quality time with it and all the omens seemed to be positive. I hope the Nuvia design group doesn't suffer QDT's fate.
The cancellation was most directly related to the attempted hostile takeover of Qualcomm by Broadcom (now being run more like a asset roll-up company than tech company) led by Hock Tan and private equity firm Silver Lake after Qualcomm share prices were crippled by their licensing dispute with Apple, which was throwing its weight around in an attempt to reduce its input costs. In an attempt to rebuff this takeover, Qualcomm pledged unfortunate cost cutting measures including scrapping Centriq and laying off most of its CPU design HQ in North Carolina to appease various shareholder advisory firms who generally guide how large mutual funds holding most of the shares will vote. In the end, that didn't sway those advisory firms, and only the US govt. invoking CFIUS powers kept an essential cellular R&D house from going on the chopping block and being sold off for parts.
It had nothing to do the technical merits of Centriq itself, and it's good to see them re-establish their once thriving CPU division in their resurgence after an almost complete legal victory over all of Apple's misleading claims. The one thing Apple did succeed in was lowering cellular licensing fees for its 5G push where it needed Qualcomm.
Do you know what happened to the NC team? Did most of them move out of state, or did someone scoop them up (like what briefly happened to AMD's Austin team with Samsung, before SARC wound down [though there is some indication people were leaving SARC long before that]).
I see QC's acquisition of Nuvia also as taking out insurance against being dependant on NVIDIA/ARM for CPU core designs. Using stock ARM IP in one's SoC has upsides, but also the principal downsides of that design being available to others (Samsung, Mediathek, others), and depending on a key competitor for that. QC and NVIDIA compete more and more in AI/deep learning acceleration, autonomous driving and related driving assistants and more. And while there are legal firewalls in place to avoid IP "leakage", it's still an uncomfortable relationship. Plus, Apple being perennially ahead in perf/W by about one generation probably chafes a bit, too. As for the question if a Nuvia-based SoC ultra portable will see the light of day, that'll depend a lot on how much effort Microsoft puts into Windows 64 bit on ARM; with enough native 64 bit software, it'd be highly competitive with x86/x64 solutions in the low and very low power space (below 10 W and below 5 W); Apple's M1 demonstrated what is possible with a ARM-based design and native OS support.
The greater risk to the development of ARM cores is from NVIDIA not buying ARM, not from NVIDIA buying it. ARM is in flux right now and so its future is unsure. It can't be assumed that ARM, let's say post IPO, is the same as ARM under SoftBank, who increased R&D beyond what an independent ARM answering directly to shareholders likely would be able to maintain. The reason Qualcomm doesn't want the merger to happen is because they don't want a (mostly future) competitor to be stronger than it otherwise would be. If Qualcomm were really worried about ARM itself under NVIDIA then they wouldn't be further committing to the ARM ecosystem by buying Nuvia.
Let me add more context to my comment. Qualcomm, if you remember, used to develop their own ARM cores. Lots of companies did. Eventually ARM's own cores took over much of the SoC market as the companies found it cost-ineffective to duplicate ARM's efforts under SoftBank. Now Qualcomm no longer has the team or the up-to-date designs it once had to develop their own cores. Seeing the possibility that they will need to begin doing it once again (because ARM is in flux and may not be investing so much in CPU core R&D as they did under SoftBank) and seeing Nuvia available, which presumably has cores that impressed them, they decided to buy Nuvia rather than start over again.
They need to be doing custom core design. Last time they did was during 820 because of the 810 disaster they had to control else it was blowing them apart add the 64bit too, 805 was a super fast CPU but it was 32bit. Then they moved all those custom designers to develop Centriq which was heralded to peak by Cloudflare. It was all axed and dissolved.
Now there's no custom design anymore, Qcomm will only do a custom core design if there's a need, right now I simply do not see the need. They always change a few aspects of the ARM reference design just like Huawei Kirin and now recent Samsung Exynos, Samsung tried but failed so many times as we know.
The reason they do not is simply because Android powered flagships do not suffer any User experience jank / poor framerates or anything vs an iPhone with far more SPEC boasted A series.
If no company comes out with an ARM SoC that can compete with Intel on desktops PCs, then I doubt things will change much in terms of native ARM software availability, which will render ARM Windows computers a small niche in the market for a long time.
I'm confused, in the prior article NUVIA was started because the founders couldn't do a server chip at Apple only laptops and phones. Now QCOM bought them and have them not do server chips but do laptops? er ok.
The company founders are getting a heck of a lot more money from Qualcomm to not make server chips than they would have got from Apple to not make server chips.
The other problem: NUVIA makes a great CPU core for servers / datacenters. Now what? Can they afford tape-out at TSMC or even Samsung? Can they convince enterprise customers to integrate a (literally) first-gen CPU? The validation costs (time, money, resources) alone might've been a bigger friction point, too. They've never sold a single CPU, much less an enterprise-class CPU.
I'm thinking of security patches, multi-year contracts, reliable supply, support & service, optimization, etc. Was NUVIA *really* ready for that? I wonder.
NUVIA would've had an excruciating battle going at it alone, I'm afraid. At least for the time being, NUVIA has a win-win. They get loads of cash, instant & essentially guaranteed capital (aka a department budget now), instant TSMC / Samsung connections & resources from an established current vendor, relative freedom to pursue a server / datacenter CPU on a longer timescale if necessary, and likely the legitimacy to attract more talent + more engineers.
That is, NUVIA was likely always going to end up either in an unusual licensing situation (a la Arm) or a very pricey // risky phase of selling new, relatively untested CPUs to trillion-dollar hyperscalers. So I almost have to wonder if this was always their plan.
Venture capital is good, but it's not necessary reliable nor enough nor patient. :(
>NUVIA’s prompt acquisition and immediate disclosure of plans to tackle the high-performance ultraportable laptop market could be seen as Qualcomm’s direct response to the new Apple M1 powered laptops and to compete with their high-performance CPU cores.
One great note: this is exciting to hear, as AMD & Intel have rather lackluster responses for the next few years (unless K12 ever does anything more than sitting in a bin). Zen4 at TSMC's 5nm, Meteor Lake at Intel 7nm: it's not going to address the fundamental perf-watt advantage x86 has. Hybrid has a small chance, as it's pairing high-power x86 with the worst cores imaginable: low-power x86. Thus, from Qualcomm & Apple, the 2022 / 2023 laptop market should be quite exciting.
2022 and 2023, that's 2 years in the future. How would you even know what is going to happen at that point of time ? AMD's Cezzane will crush every single CPU in the market for the BGA use and throw machines, be it M1 (which already got beaten by Renoir) and the future Tiger Lake, Intel is not showing any solid evidence to believe in them. Latest 10nmSF is not even being used for the new Xe iGPU, so how can I even trust them since Intel's major weakness is in the Fab engineering right now. Whereas on the other hand Zen4 is on track on TSMC 5nm, without any bullshit Biglittle (Intel's tactic to get into the SMT performance lost for the lack of Ring Bus successor for more core CPUs) for Desktop market.
Do you qualify every post with ultra-specific niches until you feel "technically correct, predominantly wrong"?
"for BGA use" = lmao "got beaten by Renoir" = Renoir that consumes 2x the power and 2x the heat? "the future Tiger Lake" = Tiger Lake H that consumes 4x the power and 4x the heat?
Apple has released two new microarchitectures every year ~for a decade~. AMD & Intel can barely release one a year and sometimes, they'll both sell you reheated uarches from last year. x86 has such slow cadence that smartphone SoC manufacturers must chuckle, especially for their performance cores.
Here's the tiny quad-core-perf 3.2 GHz // 6.3W M1 beating the 16-core 5.05 GHz // 20.6W $799 5950X in single-threaded performance:
Nobody believes Intel's future looks good. Who said that? It's clear Apple outclassed AMD & Intel & Qualcomm & Arm's stock cores; Apple has a much more serious intention to change perf-watt than the rest combined.
Apple gets TSMC's newest nodes, so that limits AMD's advantages perennially. Today, Intel has the worse nodes, Qualcomm / NVIDIA / AMD fight over the middle, and Apple gets the best nodes. The middle & worst might change, but Apple's trillions are unfortunately going to shut out anyone with mere billions for the near future.
>NUVIA’s prompt acquisition and immediate disclosure of plans to tackle the high-performance ultraportable laptop market could be seen as Qualcomm’s direct response to the new Apple M1 powered laptops and to compete with their high-performance CPU cores.
2019: A13 (Lightning) 2020: Apple M1 2021: Apple M2 2022: Apple M3 vs NUVIA Phoenix 2023: Apple M4 vs NUVIA ...
NUVIA's 2022 core is faster than Apple's 2019 core, which is exciting but also needs to be put in perspective. I hope NUVIA can iterate as rapidly as Apple: two new microarchitectures per year, one high-performance focused and one efficiency / power-focused. None of this AMD / Intel cadence of 1.5 to 3 years between new microarchitectures.
Hopefully Phoenix has its 2nd-generation successor already in deep development.
Bizarre to think that in 5-10 years, even a rejuvenated Intel might find itself in the middle of a pack with Apple, Qualcomm(/Nuvia), NVIDIA(/ARM), AMD, Amazon, and maybe some other maker like Ampere (with the Altra chips) or some future acquirer (Google?). That future isn't guaranteed, and some of these cos. probably drop out of the running, but enduring multi-way competition seems really plausible.
Reminds me of the transition from Microsoft being on top of the world to the constellation of tech companies we have now. Wonder what new problems would replace today's--chips tied to the rest of a company's platform, maybe? Maybe the few leading foundries find themselves in a powerful position when all the fabless chipmakers are competing with each other? We'll see, I guess.
Intel will (presumably) still have their own fabs though. Even if hell freezes over and they switch to ARM, that's a significant advantage, especially in chiplet designs, as it mitigates the bloody bidding war over TSMC/Samsung capacity.
Yeah, platform silos are already a problem, and I can see it spreading.
The cost to change a fab to a new process where everything is different, is about the same cost as just building a new fab. So owning what they have can be considered a hinderance unless they continue on legacy product.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
25 Comments
Back to Article
SarahKerrigan - Tuesday, March 16, 2021 - link
I was really excited about Centriq - talked to some people who had spent quality time with it and all the omens seemed to be positive. I hope the Nuvia design group doesn't suffer QDT's fate.Raqia - Tuesday, March 16, 2021 - link
The cancellation was most directly related to the attempted hostile takeover of Qualcomm by Broadcom (now being run more like a asset roll-up company than tech company) led by Hock Tan and private equity firm Silver Lake after Qualcomm share prices were crippled by their licensing dispute with Apple, which was throwing its weight around in an attempt to reduce its input costs. In an attempt to rebuff this takeover, Qualcomm pledged unfortunate cost cutting measures including scrapping Centriq and laying off most of its CPU design HQ in North Carolina to appease various shareholder advisory firms who generally guide how large mutual funds holding most of the shares will vote. In the end, that didn't sway those advisory firms, and only the US govt. invoking CFIUS powers kept an essential cellular R&D house from going on the chopping block and being sold off for parts.It had nothing to do the technical merits of Centriq itself, and it's good to see them re-establish their once thriving CPU division in their resurgence after an almost complete legal victory over all of Apple's misleading claims. The one thing Apple did succeed in was lowering cellular licensing fees for its 5G push where it needed Qualcomm.
jeremyshaw - Tuesday, March 16, 2021 - link
Do you know what happened to the NC team? Did most of them move out of state, or did someone scoop them up (like what briefly happened to AMD's Austin team with Samsung, before SARC wound down [though there is some indication people were leaving SARC long before that]).cpuaddicted - Tuesday, March 16, 2021 - link
Microsoft hired most of them to work on their ARM server processor.eastcoast_pete - Tuesday, March 16, 2021 - link
I see QC's acquisition of Nuvia also as taking out insurance against being dependant on NVIDIA/ARM for CPU core designs. Using stock ARM IP in one's SoC has upsides, but also the principal downsides of that design being available to others (Samsung, Mediathek, others), and depending on a key competitor for that. QC and NVIDIA compete more and more in AI/deep learning acceleration, autonomous driving and related driving assistants and more. And while there are legal firewalls in place to avoid IP "leakage", it's still an uncomfortable relationship. Plus, Apple being perennially ahead in perf/W by about one generation probably chafes a bit, too.As for the question if a Nuvia-based SoC ultra portable will see the light of day, that'll depend a lot on how much effort Microsoft puts into Windows 64 bit on ARM; with enough native 64 bit software, it'd be highly competitive with x86/x64 solutions in the low and very low power space (below 10 W and below 5 W); Apple's M1 demonstrated what is possible with a ARM-based design and native OS support.
Yojimbo - Tuesday, March 16, 2021 - link
The greater risk to the development of ARM cores is from NVIDIA not buying ARM, not from NVIDIA buying it. ARM is in flux right now and so its future is unsure. It can't be assumed that ARM, let's say post IPO, is the same as ARM under SoftBank, who increased R&D beyond what an independent ARM answering directly to shareholders likely would be able to maintain. The reason Qualcomm doesn't want the merger to happen is because they don't want a (mostly future) competitor to be stronger than it otherwise would be. If Qualcomm were really worried about ARM itself under NVIDIA then they wouldn't be further committing to the ARM ecosystem by buying Nuvia.Yojimbo - Tuesday, March 16, 2021 - link
Let me add more context to my comment. Qualcomm, if you remember, used to develop their own ARM cores. Lots of companies did. Eventually ARM's own cores took over much of the SoC market as the companies found it cost-ineffective to duplicate ARM's efforts under SoftBank. Now Qualcomm no longer has the team or the up-to-date designs it once had to develop their own cores. Seeing the possibility that they will need to begin doing it once again (because ARM is in flux and may not be investing so much in CPU core R&D as they did under SoftBank) and seeing Nuvia available, which presumably has cores that impressed them, they decided to buy Nuvia rather than start over again.Silver5urfer - Wednesday, March 17, 2021 - link
They need to be doing custom core design. Last time they did was during 820 because of the 810 disaster they had to control else it was blowing them apart add the 64bit too, 805 was a super fast CPU but it was 32bit. Then they moved all those custom designers to develop Centriq which was heralded to peak by Cloudflare. It was all axed and dissolved.Now there's no custom design anymore, Qcomm will only do a custom core design if there's a need, right now I simply do not see the need. They always change a few aspects of the ARM reference design just like Huawei Kirin and now recent Samsung Exynos, Samsung tried but failed so many times as we know.
The reason they do not is simply because Android powered flagships do not suffer any User experience jank / poor framerates or anything vs an iPhone with far more SPEC boasted A series.
Ppietra - Tuesday, March 16, 2021 - link
If no company comes out with an ARM SoC that can compete with Intel on desktops PCs, then I doubt things will change much in terms of native ARM software availability, which will render ARM Windows computers a small niche in the market for a long time.webdoctors - Tuesday, March 16, 2021 - link
Some ARM Chromebooks were supposed to, but at least Samsung's offerings there were pretty weak.webdoctors - Tuesday, March 16, 2021 - link
I'm confused, in the prior article NUVIA was started because the founders couldn't do a server chip at Apple only laptops and phones. Now QCOM bought them and have them not do server chips but do laptops? er ok.Yojimbo - Tuesday, March 16, 2021 - link
The company founders are getting a heck of a lot more money from Qualcomm to not make server chips than they would have got from Apple to not make server chips.ikjadoon - Tuesday, March 16, 2021 - link
The other problem: NUVIA makes a great CPU core for servers / datacenters. Now what? Can they afford tape-out at TSMC or even Samsung? Can they convince enterprise customers to integrate a (literally) first-gen CPU? The validation costs (time, money, resources) alone might've been a bigger friction point, too. They've never sold a single CPU, much less an enterprise-class CPU.I'm thinking of security patches, multi-year contracts, reliable supply, support & service, optimization, etc. Was NUVIA *really* ready for that? I wonder.
NUVIA would've had an excruciating battle going at it alone, I'm afraid. At least for the time being, NUVIA has a win-win. They get loads of cash, instant & essentially guaranteed capital (aka a department budget now), instant TSMC / Samsung connections & resources from an established current vendor, relative freedom to pursue a server / datacenter CPU on a longer timescale if necessary, and likely the legitimacy to attract more talent + more engineers.
That is, NUVIA was likely always going to end up either in an unusual licensing situation (a la Arm) or a very pricey // risky phase of selling new, relatively untested CPUs to trillion-dollar hyperscalers. So I almost have to wonder if this was always their plan.
Venture capital is good, but it's not necessary reliable nor enough nor patient. :(
ikjadoon - Tuesday, March 16, 2021 - link
Right on time. Qualcomm's CEO Cristiano Anon had hinted at an aggressive NUVIA timeline a week ago, pending the acqusition (interview here: https://www.cnet.com/news/from-ps5-to-ford-f-150-h... )>NUVIA’s prompt acquisition and immediate disclosure of plans to tackle the high-performance ultraportable laptop market could be seen as Qualcomm’s direct response to the new Apple M1 powered laptops and to compete with their high-performance CPU cores.
One great note: this is exciting to hear, as AMD & Intel have rather lackluster responses for the next few years (unless K12 ever does anything more than sitting in a bin). Zen4 at TSMC's 5nm, Meteor Lake at Intel 7nm: it's not going to address the fundamental perf-watt advantage x86 has. Hybrid has a small chance, as it's pairing high-power x86 with the worst cores imaginable: low-power x86. Thus, from Qualcomm & Apple, the 2022 / 2023 laptop market should be quite exciting.
ikjadoon - Tuesday, March 16, 2021 - link
*fundamental perf-watt advantage Arm has.Sigh.
ikjadoon - Tuesday, March 16, 2021 - link
*AmonA deeper sigh.
Silver5urfer - Wednesday, March 17, 2021 - link
2022 and 2023, that's 2 years in the future. How would you even know what is going to happen at that point of time ? AMD's Cezzane will crush every single CPU in the market for the BGA use and throw machines, be it M1 (which already got beaten by Renoir) and the future Tiger Lake, Intel is not showing any solid evidence to believe in them. Latest 10nmSF is not even being used for the new Xe iGPU, so how can I even trust them since Intel's major weakness is in the Fab engineering right now. Whereas on the other hand Zen4 is on track on TSMC 5nm, without any bullshit Biglittle (Intel's tactic to get into the SMT performance lost for the lack of Ring Bus successor for more core CPUs) for Desktop market.ikjadoon - Thursday, March 18, 2021 - link
Do you qualify every post with ultra-specific niches until you feel "technically correct, predominantly wrong"?"for BGA use" = lmao
"got beaten by Renoir" = Renoir that consumes 2x the power and 2x the heat?
"the future Tiger Lake" = Tiger Lake H that consumes 4x the power and 4x the heat?
Apple has released two new microarchitectures every year ~for a decade~. AMD & Intel can barely release one a year and sometimes, they'll both sell you reheated uarches from last year. x86 has such slow cadence that smartphone SoC manufacturers must chuckle, especially for their performance cores.
Here's the tiny quad-core-perf 3.2 GHz // 6.3W M1 beating the 16-core 5.05 GHz // 20.6W $799 5950X in single-threaded performance:
https://www.anandtech.com/show/16252/mac-mini-appl...
Nobody believes Intel's future looks good. Who said that? It's clear Apple outclassed AMD & Intel & Qualcomm & Arm's stock cores; Apple has a much more serious intention to change perf-watt than the rest combined.
Apple gets TSMC's newest nodes, so that limits AMD's advantages perennially. Today, Intel has the worse nodes, Qualcomm / NVIDIA / AMD fight over the middle, and Apple gets the best nodes. The middle & worst might change, but Apple's trillions are unfortunately going to shut out anyone with mere billions for the near future.
Dug - Thursday, March 18, 2021 - link
You first claim how would you even know what's going to happen in the future, and then go on to make predictions. Great.ikjadoon - Tuesday, March 16, 2021 - link
>NUVIA’s prompt acquisition and immediate disclosure of plans to tackle the high-performance ultraportable laptop market could be seen as Qualcomm’s direct response to the new Apple M1 powered laptops and to compete with their high-performance CPU cores.2019: A13 (Lightning)
2020: Apple M1
2021: Apple M2
2022: Apple M3 vs NUVIA Phoenix
2023: Apple M4 vs NUVIA ...
NUVIA's 2022 core is faster than Apple's 2019 core, which is exciting but also needs to be put in perspective. I hope NUVIA can iterate as rapidly as Apple: two new microarchitectures per year, one high-performance focused and one efficiency / power-focused. None of this AMD / Intel cadence of 1.5 to 3 years between new microarchitectures.
Hopefully Phoenix has its 2nd-generation successor already in deep development.
twotwotwo - Tuesday, March 16, 2021 - link
Bizarre to think that in 5-10 years, even a rejuvenated Intel might find itself in the middle of a pack with Apple, Qualcomm(/Nuvia), NVIDIA(/ARM), AMD, Amazon, and maybe some other maker like Ampere (with the Altra chips) or some future acquirer (Google?). That future isn't guaranteed, and some of these cos. probably drop out of the running, but enduring multi-way competition seems really plausible.Reminds me of the transition from Microsoft being on top of the world to the constellation of tech companies we have now. Wonder what new problems would replace today's--chips tied to the rest of a company's platform, maybe? Maybe the few leading foundries find themselves in a powerful position when all the fabless chipmakers are competing with each other? We'll see, I guess.
brucethemoose - Tuesday, March 16, 2021 - link
Intel will (presumably) still have their own fabs though. Even if hell freezes over and they switch to ARM, that's a significant advantage, especially in chiplet designs, as it mitigates the bloody bidding war over TSMC/Samsung capacity.Yeah, platform silos are already a problem, and I can see it spreading.
Dug - Thursday, March 18, 2021 - link
The cost to change a fab to a new process where everything is different, is about the same cost as just building a new fab.So owning what they have can be considered a hinderance unless they continue on legacy product.
webdoctors - Wednesday, March 17, 2021 - link
Lots of server chips still out there, SPARC M8 processor and also the IBM Power9/10.schl - Tuesday, March 16, 2021 - link
Microsoft picked up the NC team and is working on building custom server CPU & AI chips under the pretext of quantum computing.https://www.zdnet.com/article/microsoft-is-hiring-...
https://www.bloomberg.com/news/articles/2019-10-02...