Microsoft is Reportedly Designing Its Own ARM Chips

Posted on December 18, 2020 by Paul Thurrott in Microsoft, Microsoft Surface, Windows 10 with 70 Comments

A new report claims that Microsoft is now designing ARM chipsets in-house that will land in Surface PCs and its cloud-based servers.

There’s not much to the report so far—it’s just two paragraphs of text—but it comes from a reliable source, Bloomberg News.

That’s OK, the details are easy enough to summarize: Microsoft is designing ARM-based chipsets in-house, as opposed to the work it’s done in the past in partnership with Qualcomm. There are at least two families of chipsets, one of which will end up in some Surface PCs and one for the cloud-based servers in its datacenters.

Microsoft previously partnered with Qualcomm on two generations of Snapdragon-based chipsets that the software giant branded as SQ1 and SQ2. And it has partnered with AMD on a customized x86-based Ryzen 3 processor for the Surface Laptop 3.

Tagged with

Join the discussion!

BECOME A THURROTT MEMBER:

Don't have a login but want to join the conversation? Become a Thurrott Premium or Basic User to participate

Register
Comments (76)

76 responses to “Microsoft is Reportedly Designing Its Own ARM Chips”

  1. derekaw

    Of course they are.

  2. sevenacids

    I don't understand the buzz about ARM. Its only advantage over x86/x64 is lower energy consumption for casual computing that leads to longer battery life and fanless designs (to a fairly limited degree) on consumer machines. But I think it's questionable that they hold up or have any benefit when it comes to compute-intensive applications. Architecture-wise, there is not much to gain. Modern x86/x64 designs, especially from AMD, are far from being legacy. IMHO, it's nothing but an industry hype and an attempt of the big players to get in control over everything - once they own the software, hardware, and infrastructure, there is no way to avoid them. Not a bright future for an open IT, I think.

    • Oreo

      In reply to sevenacids:

      Apple’s slowest ARM chips are beating the pants off of any Intel x86 chips in single core benchmarks. Only AMD’s Zen 3 cores are competitive, but they have a significant power disadvantage — which means that once Apple introduces many-core chips, also they will be significantly faster. Power efficiency and power consumption are, when you get to many-core chips, contingent on one another.


      Another important aspect is that neither Intel nor AMD at present integrate hardware accelerators, which will become important in the future, too.


      So yeah, both, the x86 architecture and their business model are legacy. To be worried about “open IT” is quite weird, because ARM is much more open than Intel with tons of SoC vendors. It is a much more vibrant ecosystem than x86 with just two players.

      • bluvg

        In reply to Oreo:

        "Apple’s slowest ARM chips are beating the pants off of any Intel x86 chips in single core benchmarks."


        PC World showed the M1 and i7-1185G7 "basically tied" in their single-threaded Cinebench test. https: //www.pcworld.com/article/3600897/tested-how-apples-m1-chip-performs-against-intel-11th-gen-and-amd-ryzen-4000.html?page=2

      • Paul Thurrott

        Winning a benchmark doesn’t mean a freaking thing. Real-world use is far more telling, and that is much more nuanced, from both performance and compatibility perspectives.
    • Greg Green

      In reply to sevenacids:

      Apple’s first arm chip in MacBook Airs performs almost as well as the high end intel chip in MacBook Pros and Mac Pros. And that’s apple’s intro low end chip.


      Also look at performance graph over time, intel vs apple arm chips. Only 5 or 6 years ago arm chips were half the speed or less of intel chips. The arm chip line has now intersected the intel chip line, and at least for the short term apple chips will improve at two or three times the rate of intel chips.


      unless apple chips run into major problems, or intel gets out of its major problems, the future for performance and efficiency is arm.

    • glenn8878

      In reply to sevenacids:

      When you consider the power advantage of ARM compared with x86, how much more performance will you get out of it? Imagine a laptop that uses the same power of a smartphone and requires one hour to charge with the same performance of a gaming platform. Even if x86 keeps up performance wise, it’s already showing strain. Intel is barely improving performance year after year. ARM has an enormous ability to scale up.

    • Paul Thurrott

      "I don't understand the buzz about ARM. Its only advantage over x86/x64 is lower energy consumption for casual computing that leads to longer battery life and fanless designs (to a fairly limited degree) on consumer machines." Yes, that's why there's buzz. :)
  3. shark47

    The article is updated and now says it’s more likely to result in server chips than in Surface chips. Also, the design unit is under the Azure business.


    Overall, though, it looks like Intel held back the industry by at least a couple years and Apple was able to call the company out. It’s remarkable, really, how a V1 product has disrupted the industry so much.

  4. jfgordon

    Let us see... what happened in the past when Microsoft tried to follow Apple's strategy: the Zune, the Kin, Windows Phone, Groove Music, the Band... Yeah, this is totally going to work out all right. /s

  5. bluvg

    In reply to sammyg:

    Fair point. :) When Microsoft has a hit, though, everything starts to revolve around it. If they start powering WVD with ARM nodes for cost-performance benefits, there still may be a consumer angle here--if they perceive it as a hit.

  6. winner

    Microsoft always chasing Apple.

    You'd think they'd originate some of these great ideas themselves, but then I'm still waiting for some consistency in the Windows 10 UI.

  7. straker135

    If industry were to decide to move to ARM for their hardware and software needs why would they stick with Microsoft as a platform vendor, and not instead go to Apple? The apparent efficiency and performance of Apple's M1 chips, the all-in approach indicating further improvements in the future surely make a compelling case for a platform shift, if an industry has to move to ARM based systems? The expense and retraining are a factor in price-sensitive industries, and they have shown a reluctance to even step up to Window 8 or 10, but if they're going to pull the trigger and have to spend heaps of cash, why not?

    • straker135

      In reply to straker135:

      Thanks for the thoughtful replies. I am reassured that the costs versus benefits remain firmly in favour of Microsoft's platform(s) and long term commitment to support in the business space. I note that I have had 10-11 year old PCs take Windows 10 (not particularly fast of course) whereas I don't think 5-6 year old iPhones/iPads get the latest operating system updates (ARM model).

    • SvenJ

      In reply to straker135: Because Apple isn't selling their chips. This is not to be greedy. They have no interest in being a processor OEM.


    • james.h.robinson

      In reply to straker135:

      Plus, what kind of track record does Apple have with the enterprise? I'm not saying their track record is bad or good, I'm just wondering what it is.

      • wright_is

        In reply to james.h.robinson:

        Generally pretty poor. They don't have the remote management and configuration tools that Microsoft provides as standard. They don't support their hardware and OSes and software for long periods of time, you have to move when Apple says you move.

        If you have an ancient piece of software that is critical to your business, there is a chance it will run on modern hardware and a modern OS. If not, you can put an older, non-supported OS on the hardware, it might not take full advantage of all the modern hardware, because drivers are missing, but it will usually boot and your software will run - even if you have to isolate the machine from the network.

        If you buy a modern Mac, you can't install an older version of macOS on it, let alone get pre-OS X software running on it.

        It is possible to add the configuration tools to macOS, but it isn't as easy or as integrated as it is under Windows.

    • Paul Thurrott

      Because there are no real-world advantages to moving to a new platform. Compatibility comes first. Familiarity. And then performance and battery life would need to be exponential improvements. Which they're not.
    • wright_is

      In reply to straker135:

      Because Apple has a poor track record, when it comes to long-term support. You move to what Apple currently deems supportable.

      With Microsoft, you can stick with what you have/need and even pay them extra for support. They also have a large range of management tools, which are generally missing from Apple's repertoire.

      Just look at Apple's history, software from 10 years ago often won't run on current hardware. Industry is often running software that is 20 - 30 years old. You can't even buy a current Mac that will run software of that vintage!

      That is Microsoft's biggest strength and its biggest weakness.

      (We have lab equipment that is still working fine, but the software is limited to Windows XP. The machines are isolated from the rest of the network and have no internet access. To upgrade them to Windows 10, we would have to throw out those working machines (over $100,000 each) to get new machines that work with software and drivers that are Windows 10 compatible.)

      Price is another consideration. We currently use PCs that come in at around 500€ per PC. The Mac mini starts at nearly double that here. It has less remote support capabilities and a lot of the software we have is Windows and Intel only, so won't even run on the Mac, and there are no Mac equivalents available. We'd have to pay to have the software re-written for WoA (move to Windows on ARM) or for the Mac. That would quickly run into 6 - 7 figures, a cost that would make the move unfeasabile.

  8. Scsekaran

    In reply to lvthunder:

    Thanks. I actually meant comparison between Surface Pro X and M1 Macbook Air with native ARM64 beta Photoshop app.

  9. madthinus

    Designing chips in-house could simply be Microsoft arranging and configuring various ARM reference designs into a soc and calling it a day. Or it could be a full fledge work up like some of the other ARM OEM have done. Both could be successful, both could be in unaspiring.

    • wright_is

      In reply to madthinus:

      There would be no point to that. The reference designs are slow, slower even than Qualcomm's chips and there are dozens of companies offering cheap, reference designs already. The only reason for designing in house is because you can't get the performance you need elsewhere.

      Although for the server side, I would probably look at Fujitsu first, as they have just set a new world record for the fastest super computer, being over twice as fast as the 2nd place IBM x64 based computer. I think their 48-core chip has a lot of potential for the Azure cloud.

  10. glenn8878

    In reply to sammyg:

    Even before M1, Qualcomm’s ARM PC chips are terribly inadequate.

  11. bart

    Wouldn't it require Microsoft to buy some company or assets of a company, to start designing an ARM-chip?!

    • Scsekaran

      In reply to Bart:

      It is a long game. The best option at the moment is Qualcomm 888 equivalent of PC chip for low powered, connected mobile chip and AMD/Intel chips for power & performance.


      How the roles have reversed!! Microsoft with predominantly x86 processor based OS to needing both ARM & X86 to cover wide variety of devices while apple moving towards M1 architecture over next couple of years from mobile to desktop unless AMD/Intel delivers a significantly powerful processor than M1

    • sergeluca

      In reply to Bart:

      Well, it took 10 years to Apple to create M1 with an ARM license.

      • SvenJ

        In reply to sergeluca: That's not really how it should be thought of. Apple didn't decide they were going to build an ARM Mac 10 years ago and focus on that. That's like saying it took MS 30 years to design Windows 10. Apple has been designing and using ARM processors in iPhones and iPads over the years, iteratively perfecting their craft.


    • ecumenical

      In reply to Bart:


      Why? All they need is the appropriate ARM license.

      • wright_is

        In reply to ecumenical:

        The ARM license won't bring them anything useful, on its own. They need the expertise of a large number of chip designers to actually take that ARM license and turn it into something efficient enough to make it worthwhile bringing out as a laptop or desktop chip.

        As sergeluca says, it took Apple nearly a decade to get from the first iPhone chips to something that is competitive in laptops.

        Even Qualcomm is only a lightly optimized version of ARM, and you can see the performance deficit that had, before Apple released their M1 and really put them in the shade.

        This isn't a market, where you can buy the license and be competitive in a few months, it will probably take 3 or 4 generations of chips, before they get anywhere near being competitive with Intel and Apple silicon, it they are lucky.

  12. Username

    What can Microsoft do independently that they can’t do better with Qualcomm? Is Qualcomm undermining potential windfall from WoA?

    • shameer_mulji

      In reply to Username:

      MS doing their own custom processors for their Azure servers makes perfect sense. As far as client side goes, I personally think they're better off partnering with AMD. They're making great x86 processors with great performance / watt.

    • wright_is

      In reply to Username:

      The market is too small for Qualcomm to really get excited about and make the necessary investment in order to make it a success. They have just optimized their mobile chips to run with a little more power, but actually invest the billions required to achieve what Apple has achieved? Doubtful.

  13. Shel Dyck

    Add the likely announcement of an AMD ARM chip on Jan 12 and things are getting really interesting.


  14. ChrisKal

    You know, I got the sense from what Chris Capossela said on Windows Weekly that this might be happening. I can't remember his exact words.


    It honestly doesn't surprise me, though. Qualcomm's efforts have been disappointing so far and I don't see any evidence that will change soon

  15. shark47

    Redmond, start your photocopiers!


    Seriously, though, didn't Mary Jo talk about this in a recent This Week in Windows episode - that Microsoft had a thousand chip designers working on server chips?

  16. IanYates82

    I suspect they'll do the same optimisation Apple did to make the x86 emulation work efficiently... Although I had assumed that's something Qualcomm would've just added to their PC-level chip


    The optimisation I'm referring to is adding a mode to the processor to enforce the stricter memory model that x86 code expects the processor to observe. Apple toggles that on and off when the processor starts and stops a block of emulated code

    • wright_is

      In reply to IanYates82:

      Qualcomm didn't invest heavily enough in the laptop chips, because the market is too small for them to bother putting many resources into. They are looking at billions of ARM chips per year for mobile devices and, maybe, a million for laptops in the next 12 months. Given that capitalism usually can't see beyond the end of its nose (i.e. the next 3 months), there is no reason to invest heavily in such a gadfly project. And it is a project that will take several years to bear fruit. It has taken Apple years to get to where they are and Microsoft's team will have to learn the same lessons that Apple has done, before they will have something competitive.

      • Scsekaran

        In reply to wright_is:

        That kind of short-sighted capitalism is one of the downfall of intel. When there was no competition form AMD and performance from ARM was not that great, they were milking consumers and businesses with as minimal advance in technology as possible. For Intel, Moore's law was considered not applicable anymore (but probably still applicable to Apple Silicon) and Intel moved away from Tick-Tock cycle.

  17. bluvg

    So that's where my forum post went ?

  18. mattbg

    Frustrating lack of details in the original report. As well as SQ1/SQ2 they also have a presumably-custom CPU in the HoloLens 2 and they have partnered with other companies to develop AI-oriented CPUs for Azure, so it seems like a natural progression.


    Microsoft's PC hardware partners hopefully have more detail on what this means because I'd have a lot of questions if I was them.

    • wright_is

      In reply to mattbg:

      I think this can only succeed if Microsoft makes the SoC or its design available to its OEMs.

      Either it is purely an Azure move and they will build custom servers for Azure, or they will need to license it to OEMs. If they don't it will cause a major split in the market. It could be the thing that breaks the Windows desktop monopoly, if they play it wrong.

  19. toukale

    Everyone needs to put a stop to this here is a quote from Panay " Microsoft plans to continue its expansion into chips, but is unlikely to focus on general-purpose chips like those from Intel, AMD and Qualcomm." This does not sound like what everyone thinks it is.

    • Greg Green

      In reply to toukale:

      What it sounds like to me is that he’s interested in special purpose chips, like what apple has for it’s os, only for windows. So far MS’ efforts at general purpose chips from Qualcomm for windows hasn’t worked. If they don’t go to special purpose arm then they’re stuck with x86.

    • bluvg

      In reply to toukale:

      Just to play devil's advocate, though, that was pre-M1, right? Microsoft tends to hedge bets. People didn't really expect the Surface line, either, but the OEMs had been spiraling downward as they churned out a steady stream of meh. If Intel is Titanic and Qualcomm the Britannic, Microsoft won't be left without a ticket for yet another liner in its pocket.

  20. codymesh

    bring the heat directly to Apple.

  21. SWCetacean

    A lot of reporters and pundits out there are positing that this is in competition to Apple's M1 desktop Arm CPUs. I really don't think that's the case, at least not in the next few years, and I don't think it's going to be a fully in-house microarchitecture like the Apple ones. To me, this looks like Microsoft copying Amazon, not Apple. Amazon has started deploying its semi-custom Graviton2 Arm server processors in AWS instances and customers have been able to use them for the past few months (AWS Graviton - Amazon Web Services, AWS Deploys Synopsys VCS on Arm-based AWS Graviton2 to Accelerate SoC Development - MarketWatch, Powering .NET 5 with AWS Graviton2: Benchmarks | AWS Compute Blog (amazon.com) ). The Graviton2 is based on Arm's own Neoverse N-1 microarchitecture, but with parts of it customized by Amazon (e.g. the Graviton2 only has 32 MB of L3 cache compared with Arm's reference design having 64 MB). Another Arm server company called Ampere has recently released reference boards with their Altra processor, also based on Neoverse N-1 but with 80 cores rather than 64 and at a higher performance level compared to Graviton2. Arm has also announced their next-generation Neoverse V-1 and N-2 microarchitectures that will be available to partners. I'm guessing that Microsoft is going to develop their own Neoverse-based Arm CPUs for deployment in Azure to compete with the lower-cost Amazon AWS Graviton2 instances.


    That's the only way I see Microsoft being able to release their own Arm chips in the next few years. It won't be a ground-up, clean-sheet design like the M1; full CPU microarchitectures take roughly 6 years to design from start to finish. It will probably be a semi-custom design that uses a Neoverse core and tunes it to Microsoft's use cases. And Microsoft has plenty of experience both with semi-custom silicon (Qualcomm's SQ1/2 and the Xbox SoCs) and Arm servers (they started deploying Marvell ThunderX2 units for internal workloads in Azure about a year ago (Company - Newsroom - Marvell’s ThunderX2 Solution Now Deployed for Microsoft Azure Development - Marvell).

    • a_lurker

      In reply to SWCetacean:

      The pundits think MS should mimic Apple but forget Apple is a hardware manufacturer first and MS is a software house first. Apple has complete control of the hardware its software runs on MS does not. Thus Apple can switch to a different CPU relatively easily while MS is more reliant on what the OEMs do.


      In house servers for Azure, Google, and AWS are often custom designs so designing your own CPU for them does make some sense. But this is a far cry for retail boxes sold to the public

    • toukale

      In reply to SWCetacean:

      I agree, that Bloomberg article was piss-poor and everyone is so desperate for an answer to Apple's M1 they are running with it. Panos already said that is not the case and his explanation matches with what you outlined with Amazon.

  22. glenn8878

    Microsoft has the billions to make its own ARM chips. The only question is why it took so long after Qualcomm failed so completely. Microsoft also failed so spectacularly with Windows RT, Windows Phone, Windows Mobile, Windows 8, and Windows on ARM. Nonetheless, they can only look forward and not past failures. Maybe for once Windows can have snappy performance. It performs so sluggishly.

    • toukale

      In reply to glenn8878:

      Having the billions does not mean you venture into areas that has no chance of a ROI. I don't see Microsoft taken the Apple path, they can't, they don't have the scale to be able to pull it off, not with the surface sales numbers. There are only two companies designing state of the art ARM chips (Apple and ARM) everyone else are using off the shelf ARM design (Qualcomm, Samsung, MediaTek). There is a reason for it, the ROI is not worth it and that won't change also in Microsoft case.

      • glenn8878

        In reply to toukale:

        ARM is hardly designing state of the art ARM chips. Their reference designs are adequate, but not as competitive as Apple. Microsoft should move faster to improve ARM reference designs. Moving faster and not relying on other vendors who have other priorities is why Microsoft is lagging behind. Microsoft is selling billions of dollars of Surface computers. That’s where the ROI matters. At some point, maybe everyone else can catch up.

  23. bluvg

    General purpose compute is general purpose compute. (And they already have FPGA and GPU specialty hardware.) It may say this is mainly targeted at Azure, but if they produce a performance champ in the datacenter, I have little doubt they would find ways to capitalize on it in the consumer market.

  24. toukale

    I don't see anyone having any ARM based chip that can compete with the M1/M1X+ for another 18-24 months at the earliest. The other issue all those companies are going to have is scale. Right now Apple is doing things to their chips which are expensive, but are able to do so because of scale. Those Mac ARM chips are funded by the iPhone A-series chips and costs is not an issue can anyone spend that kind of money when they don't have the scale or be able to ship them in devices non-Apple users will pay? Those questions tells me Microsoft will get some ARM designs and tweak it a bit and see how it goes.

  25. sykeward

    Maybe...the Bloomberg article says "Microsoft’s efforts are more likely to result in a server chip than one for its Surface devices, though the latter is possible, said one of the people", which doesn't sound like anything is coming to Surface in the near future from this effort.


    It sounds like AMD already has prototypes up and running of a desktop-class ARM processor aimed at the Apple M1, and I think that one is more likely to appear in Surface and other WoA devices. I'm interested to see what it brings to the table in any case.

    • SYNERDATA

      In reply to Sykeward:


      I have put off buying SPX1 (SQ2) because I would now prefer an SPX2 with a 12 core AMD ARM64 SoC with 4 efficiency cores and Radeon GPU and 16GB on the SoC wafer.

    • toukale

      In reply to Sykeward:

      Agreed, I don't see Microsoft taken the Apple route, it would be a multi-year expensive ventures when they don't have the scale nor the consumer base that would make such an investment worth it. What I can see Microsoft if they do decide to do so is to get an ARM design from ARM and tweak it somewhat to what they want.

  26. ponsaelius

    Everything I have read and seen about Windows on Arm (Windows on Qualcomm) makes me feel that at best it has limited appeal. At worst it is an experiment that customers are paying a lot of money for. I don't see the point of Windows on Arm while it delivers so little. A modest improvement on battery life with the issue of performance and lack of compatibility with Windows applications.


    I use Windows every day and I could not see myself using Windows on Arm. Maybe that's my limited imagination but it doesn't even seem to drive enthusiasm in Microsoft. It seems an experiment that they have been playing with for a decade and still get stuck on performance and compatibility.


    Apple, with their M1 chip, seem exactly the opposite. All in on ARM. This is where they are taking their PCs and their OS.


    If Microsoft are going to continue to address ARM as a lukewarm experiment that dips a small toe into the water, then no one will take it seriously. it will be interesting. It will be for fans and early adopters. The people who bought WindowsPhone, had Groove Music, a Band, used Cortana, etc. A bit like me, except for the Band because I am in the UK. Microsoft have got people to buy in to visions in the past and have burned bridges and trust. It makes selling ARM more difficult.

    • Scsekaran

      In reply to ponsaelius:

      It certainly has limited appeal for now but I think it is born out of necessity rather than experiment. In a slim surface form factor with cellular connectivity/Instant On and a decent battery life, It is impossible with x86 processors either from intel or AMD currently. The battery life in Surface Pro 7 is atrocious even without LTE connectivity. Qualcomm 8cx processor in a clamshell form factor- Samsung Galaxy book S has battery life close to 20 hours with LTE. The performance would be closer to Core i5 in native apps but performance and battery life will be significantly degraded in emulation(but at least there will be some decent compatibility).


      In the current situation, Microsoft has to rely on intel or AMD for power and performance but Qualcomm for thin/connected devices with decent battery life with decent performance in native apps. The key is native apps rather than emulated and persisting.


      Apple M1 has changed the landscape. I am yet to see any benchmarks/comparisons of native photoshop version on M1 vs. Surface Pro X 2 but I wouldn't hold my breath.

    • wright_is

      In reply to ponsaelius:

      ARM has a lot of scope for use in datacentres. If it becomes popular there, it could filter back down into desktop and laptops. The performance and power usage advantages of ARM over Intel are undeniable.

      But, until Apple this year, nobody really invested in the use of ARM in laptops or on desktops.

      There were hobbyist devices, like the Raspberry Pi, but the laptop and desktop markets are so small, compared to mobile, that nobody was willing to invest in it, at least not seriously.

      The Qualcomm chips were mildly tuned mobile chipsets and it shows in their laptop performance - lacklustre.

      Apple has now shown that there is a case for ARM in laptops and, hopefully desktops - we will have to see what next year brings, before we have an answer there. But Apple is all-in on ARM.

      If Microsoft can show the same dedication to ARM that Apple has done, it stands a chance. But they need to bring the developers with them. As long as the software remains purely x86 and x64, ARM won't stand a chance. And it is legacy software, especially legacy drivers and software for industrial devices (production lines, laboratory equipment, like spectrophotometers etc.) that will need to come forward, but the manufacturers of these devices have traditionally abandoned older hardware, preferring to sell a new $100,000+ peripheral to get drivers for a newer version of Windows. Who in their right mind is going to give out over $100K for drivers for a $100 part (Windows)? Which is why a lot of production lines and laboratories still have extensive fleets of XP and Windows 7 PCs, because there is no other option.

      And if a corporate has to keep a fleet of x86 and x64 machines running, they will probably stick with Intel for everything, making support easier. There will need to be an Apple like performance improvement to make corporate buyers sit up and take notice and, maybe, abandon Intel for ARM for "normal" users going forward.

      The industry certainly needs this shake up, the question is, whether Microsoft has the will to see it through, this time.

    • iPhoneX

      In reply to ponsaelius:


      You said it yourself - it is a matter of imagination.


      This may be another RT or Zune, but hopefully this bandwagon pays off.


      Fingers crossed.

  27. crunchyfrog

    This makes more sense than Microsoft's current strategy. More control of the hardware and software will give them the edge they've been lacking. Now that they have seen what can be possible with in house designs, I have no doubt that we'll see vast improvements in performance.

Leave a Reply