Microsoft Designing Its Own Chips for Servers, Surface PCs
bloomberg.com
bloomberg.com
Also, given they sell an OS to other manufactures to install on their own machines, they are going to keep that running for a while. Same with servers... Even in the Azure side, they are allowing enterprise and OEMs build Azure Stack. I cant see them locking it down... But Apple? yes... its already locked down...
The thing to worry about is not them "locking it down" it's that they're building custom hardware that you can only access by renting cloud computing (EC2) or buying a complete machine (Apple).
I doubt Microsoft will lock out Intel or AMD, but we could see certain Windows features compatible only with Microsoft silicon.
I don't think Amazon will throw in the resources to build a chip just to offer cheaper prices than competitors alone. Pretend you are Amazon. Remember, the cost of building a chip is similar between you and another guy. And if you are getting less return on investment compared the other guy why bother? It has to be that getting into chips will give you more return on investment than the other guy. But return on investment comes from many forms. For example, using graviton to run VMs is cheaper and leading your customer to use AWS more and less likely to leave AWS is another form of return. In this strategy, you need to build up the barrier for customers to leave. You could do it by making the stuff customer use with their VMs have lock-in effect, for example databases or storage that only runs or works better on AWS. Or you can directly make sure things customers run on Graviton cannot be run on anywhere else.
By the way, a better hardware performance is hard to maintain. If everyone uses the same instruction sets and the same manufacturing process, the only way you can be faster than the other guy by better circuit design. But there is only so much you can squeeze out of a circuit design and its really hard to maintain a sizable lead for many many years. Look at how AMD designed a better circuit and Intel's competitive position is gone. Changing the way software interact with the hardware is next easiest thing to do, and especially easy if you have a lot of leverage over software, OS, APIs, programming languages and developers. Basically what companies like Amazon, Microsoft, Google and Apple all excel at. See the addition of matrix operations for ML significantly increased performance. And how Apple can use their control over iOS APIs to build CoreML, get developers to use it, and have add special instructions and function blocks to their hardware to accelerate core ml. It will be interesting to see how this develops, but it would be only natural for these companies to explore this direction.
That pretty much sealed the deal for any serious laptop makers and now we've got Microsoft embracing "alternate" instruction sets, Amazon embracing non-Intel data center architectures, and Apple out competing with a bespoke CPU in a bespoke laptop form factor.
<opinion!> Intel is toast. (if you were still wondering) Which truly sucks because you really really want and need a chip fab business that can compete on the global scale. Intel needs to be that business for the US (Sorry Global Foundries you had your chance)
RISC-V is the new 8080A, FPGAs are the new SoC. Look for a huge number of computers available in all price points based on RISC-V sitting inside FPGA fabrics. This is going to be driven by open source tools and a set of open source HDL cores for all of the "essential" peripherals.
The consumer device kingmakers will be graphics architecture suppliers. The biggest differentiator for the next few years will be "can your graphics card do this?!" </opinion!>
Not really. All this shuffling to "custom hardware" is due to the fact that Moore's Law broke a couple generations back.
Custom hardware makes no sense if you can just wait 18 months and the performance of a generic chip will double.
Once you can't just "wait for performance increases" then building your own silicon starts to make sense.
Custom hardware makes no sense if you can just wait 18 months and the performance of a generic chip will double. Once you can't just "wait for performance increases" then building your own silicon starts to make sense.
As a restatement of the economics of building competitive systems.
My comment was that the economic forces that allowed IBM, HP, and DEC to build custom hardware and recover those costs through sales have returned. The mechanism of that return is that Apple has shown that product differentiation through custom silicon is sufficient to add margin to a product above and beyond the cost of developing that custom silicon.
Your comment appears to say the same thing; which is that the previous strategy of maximizing margins of letting a semiconductor company produce a better chip and then use it, is no longer more effective than you telling a semiconductor fab to make the chip you want to use and then using that.
That is what you are saying isn't it?
Crystal ball tells me that someone will likely be Microsoft. If Microsoft has high performance Surface chips, there is little in the way of them licensing it to OEM partners. Intel, AMD, Qualcomm, etc... won't be able to compete with MS, due to the vertical integration between the Windows operating system and Microsoft silicon.
(Incidentally, this is why I see efforts to remove backwards-compatibility from the PC as a hostile takeover and closing of the ecosystem.)
https://en.wikipedia.org/wiki/WHQL_Testing
The only way you get a standard like that is if a really big player, in this case Microsoft, steps in and sets the rules and does the work. It's not like backwards compatibility had enough inertia to do this without having someone set the standard.
No, the PC/AT was the de-facto standard and that came long before WHQL or anything else from MS.
Just look how hard it is to migrate a significant application from Oracle to SQL Server or Postgres. It's easy for a trivial application, but there are lots of things to trip you up if you're migrating some big hairy complicated application.
I'm not sure how much we will see divergence, or how much it matters to most developers. If the tools can compile/target a given CPU and operate, devs largely won't care. Some will, most won't.
I don't think that MS and Amazon will stray too far from generally supported 64-bit arm instructions as a base. Maybe hardware optimizations that will work better in some workloads, but the baseline will likely work. This is only because of the diverse languages and tools that they need to support.
I think it's more about shoring up options, reducing costs and optimizing data center layouts. When targeting cloud, many apps are already scaling horizontally. So you don't need to perform like an M1 for general use as long as you can scale horizontally enough at a competitive price point, you can get a very long way without a ton of custom extension.
Microsoft have considered their arm desktops as less serious than the others? If this new arm chip is meant to be an intel challenger that lower status must end. Universal binaries solve that too.
Imagine this: stick an M1-like CPU and GPU combo on a PCIe card. Run the games entirely on the card, using the host system as merely a way to provide data to the card. In one shot, you murder the AMD and Intel processor market - everyone can just buy the cheapest model and still play games, saving their money for the unified GPU/CPU board. It also paves the way for eGPU products for laptops becoming more mainstream, and could work towards a play for the Shield to become a console competitor (if cannibalizing their existing console business is worth it to them).
Also NVIDIA quit the mobile market a while ago. They are focused on Autonomous driving and embedded market. The only consumer product on the market with NVIDIA SoC is Nintendo Switch (which sells millions)
But what if Microsoft takes the lessons learned from making their server chips, and uses that to develop their own chips for the Surface. Again, bad for Intel, but hardly a huge deal.
But what if Microsoft then takes these Surface chips, and licenses them to Windows manufacturers to create the next generation of PCs... PCs that can compete with the M1 in terms of performance and efficiency.
For Microsoft, the benefits would be immense. Microsoft obviously knows Windows better than anyone, and would be able to tailor both the operating system and their CPUs to work far more efficiently than Wintel. This also increases Microsoft’s vertical integration and control over the Windows operating system, which they’d no doubt view as positive. Microsoft becomes vendor of the Windows operating system and the CPUs it runs on. Intel systems would be at a permanent competitive disadvantage, as they don’t have the tight level of integration of Windows and MS-designed chips.
The off the shelf ARM core designs are likely good enough for Microsoft to create an offering that’s competitive with Intel, if they implement some tweaks to make it integrate and perform better with Windows (as Apple has done with M1 and macOS)
This might sound bonkers right now, but there’s definitely a version of the future where we’re buying a Dell XPS powered by Windows 10 and “Microsoft Silicon”
This is all happening because ARM licenses it’s core design and ISA to anyone. It’s becoming relatively easy for major internet companies to design and implement their own bespoke CPU designs, optimized for their unique workloads. This means better performance, and lower energy costs, which is critical for these massive datacentres. The optimization benefits offered by that degree of vertical integration is something Intel may not be able to compete with.
For Intel this is obviously terrible news. Severs are their second biggest cash cow, and MS might end up being merely among the first of many internet companies to pursue their own CPU designs. This could foretell a rapid shift away from Intel and x86 in the server space.
This has implications for their PC chip business too. More money invested into ARM processors, means more R&D for ARM-based designs, which means better ARM chips in PC and mobile. This makes Intel less competitive in the PC space.
However, it would be a terrible move for interoperability. OEMs would be inclined to purchase Microsoft's chips (MS's vertical integration would mean that their chips would outperform the competition when running Windows), and Microsoft could in turn restrict their chips to the Windows operating system.
How much performance do you get by taking, say, code originally written for a 10-year-old compiler whose default target was vanilla i486, and build wuuth a recently built compiler? I'd expect wins from targeting a more modern instruction set, plus 10 years of general "we've learned more about optimization".
Microsoft can't do that without an architecture change, and if they do that, they're basically abandoning their primary push to consumers: your established software and workflow investments are no longer safe on Windows.
Also, the main motivation is the cost. Any other vendor would want the highest profit out of their design.
In my opinion, the majority of people have too much faith in their ability to predict the future.
I was terrible at predicting the real world performance of the Apple Silicon M1. It is, in fact, much better than I expected. On the other hand, Microsoft has thus far only had slightly modified AMD chips in their Surface Laptop, and poor performing ARM-designed Qualcomm chips in their Surface Pro X. Maybe I'll be bad at predicting the future, but I do not expect excellent performance out of Microsoft's Surface chips in the next 365 days. Probably longer.
In the meantime, more Windows computers will be sold than MacOS, and they will have mostly Intel chips, but an increasingly large number of AMD chips.
AMD has survived with less diverse revenue streams and much worse product portfolios. I'm optimistic for how they'll do over the next several years.
I expected what came out in the end. Apple would never put up the amount of money and the promise to ditch Intel if they were not absolutely certain that they could actually beat Intel performance-wise and have a working Rosetta to keep "old" software running.
The writing was on the wall for a long time, iPhone and iPad processor power has been taking decent shots at moderately powerful PC hardware for years now - the key thing why Apple didn't do it two years ago was software support and developer tooling, they wanted to avoid cloning the Windows RT fiasco which fizzled out because no one had working software and there was no translation layer.
It's also the case that now Apple has shown what can be done more firms will be seeing how they can try to reproduce that which will spur more investment in the Arm ecosystem.
Most of their hardware products (not all... but most) end up being good, but not earth-shatteringly so, and either keep kind of moving on without taking any crowns (like Xbox), or slowly wither and die (Zune, Windows Phone).
Surface seems to be a good Halo product (not the game, ha!) so far, but I see very few Surfaces in the wild compared to any other laptop/tablet/desktops (mostly HP, some Dell, etc.).
Perhaps the closest thing to a Zune that Microsoft makes today is the Surface Duo. It's very expensive with nice hardware, but the initial launch got some bad press from mediocre software. The software has likely gotten a lot better over time, so it will be interesting to see a version 2 here. But no one will be buying a $1400 Zune :)
Microsoft must always 2nd tier any consumer effort to their business of business.
It does put healthy pressure on them, but I think AMD is fine for the near future, they're also working on ARM chips as well.
Plus, it's very likely that unlike Apple, Microsoft will not be able to find a way to economically design their own CPU cores. They'll just be premade Cortex X1 or whatever licensed ARM core, but Apple will remain ahead with Firestorm/Icestorm and whatever comes after. Kind of like Qualcomm right now.
The chips don't have to be better than Apple, they just have to be better than Intel (and AMD).
The benchmarks of the Snapdragon 888, which utilizes the Cortex-X1 core design, look decent. I mean, they're not as good as the A14 and M1 with the Firestorm/Icestorm cores, but they're decent. Decent enough that Microsoft could likely build upon the X1 core and tweak those designs to have competitive performance to Intel.
Can read about another chip they are making here: https://www.geekwire.com/2020/microsoft-pluton-new-chip-desi...
The CPU on the Xbox 360 was co-designed with IBM, working to integrate their PowerPC-based PPE cores with an AMD GPU.
As an outsider looking in, Microsoft definitely dabbles its toes into designing chips but they've always done it with a partner already big in the space of chip design. Even with that Pluton "chip", its more about creating a standard secure enclave system between existing CPU manufacturers based on some Microsoft specifications rather than an entire chip design ready to get sent to a fab.
Microsoft is just as rich as Apple. They can do whatever they want.
I highly doubt they would get into the trouble of designing their own chips only to settle for something mediocre.
If you can fit 500-1000 mediocre arm cores in the same space as 128 x86 cores at half the power/heat, it absolutely makes sense for something like Azure. They don't have to beat Intel/AMD on a core for core basis as long as they can compete on compute/space/power requirements horizontally.
It's not the best option for some loads, but is a good enough option for enough loads that the diversification can really work out. There's plenty of software/systems/services that don't need x86 per core performance and a single ARM core is enough per node.
I just wrote about that [2] few hours ago before this news pops up. Even Microsoft is abandoning the WinTel Alliance. I know there are still long way to go before the dismissal of x86 / Intel. On one hand they deserved it, for pathetic management in a hyper competitive market. On the other hand I kind of feel sorry for them. Seeing the Giant falling.
[1] https://www.servethehome.com/impact-of-marvell-thunderx3-gen...
Me too. Except they really did deserve it but not because they could not compete, but because instead of trying to help board manufacturers making better computers they sued them.
Nvidia made an awesome chipset with a mini cpu made to handle basic instructions and used intel cpu for the high end computation. Intel sued them for not using their own Chipset.
The nvidia chip was a power saver in 2010. And it was the best of both worlds. Zotac put together their chipset and board into a mini itx pc. Fabless. Playing 1080p video.
That was the height of innovation and intel tried to keep a clamp on the market for fear of losing it.
And maybe that is the bottom line. They knew their product was dying. Their market was no longer going to grow. So they followed the 40% rule. And tried to milk the market as much as they could.
Source. https://www.zotac.com/us/product/mainboards/zotac-ionitx-t-s...
Or maybe there's something about full integration that makes a custom-designed chip just so much better than a more general purpose chip?
These major strategy shifts require enormous courage, support and trust at the executive level, because they represent cannibalizing an existing (real) part of the business for the promise of future profits in a business that does not yet exist.
Kodak is a textbook example. Execs refused to pivot to digital cameras and associated products, because this would compete with the enormously profitable (and internally powerful) film business.
In hindsight, yes, it was a mistake. But it's a very understandable one.
For servers there's Ampere and Nuvia. For ARM PCs I think Qualcomm/Samsung/MediaTek/etc. could easily build a pretty good SoC if there's obvious demand.
No they couldn't. I mean, not from a technological point of view, there's nothing holding at least Qualcomm and Samsung back.
The problem is the software side. Unlike Apple, all three are used to dumping layers of bullshit upon fossilized bullshit and then dropping that off to customers hoping no one notices that the kernel version is years old, the kernel sources and the entire BSP are nowhere even close to upstream Linux quality - ever seen Mediatek leaked Android kernels? Stay far FAR away if you value your sanity, and the others aren't better. Security for all of them is an afterthought.
Apple created everything from scratch, with security in mind. I agree that Apple ain't perfect either, but... the iPhone 4S was released 2011, it took eight years until the checkm8 bootrom exploit was found. The Android world doesn't even come close.
The software for ARM PCs is Windows. All the SoC vendor has to provide is some UEFI drivers. Linux will survive on reverse engineering since ARM PCs will have no Linux BSPs, not even shitty ones.
If Intel can avoid a collapse in market share over a five year horizon they will have done very well.
What might save them is a major issue with TSMC (possibly a political one).
- AMD is as good as or beating Intel on performance on server, desktop and laptop (not just desktops).
- Apple (and now MS) looking to control CPUs in their laptop and desktop products and producing Apple has a superior product as a result.
- Nvidia will soon own Arm and will be pushing it's own Arm CPUs.
- The cloud vendors all the see the (price) advantage of Arm based CPUs and are starting to promote them.
- Even a startup like Ampere can produce a product that is competitive with Intel.
That's a lot of competition.
Add to this that Intel is still having process problems.
So on a backward looking basis all looks fine. Looking forward not so much.
And the market has exactly the same concerns as I do. Intel has a P/E ratio of less than 10 with strongly growing revenue and eps (even boring old IBM and Dell have much higher P/E ratios). Why? Because the market has no confidence it can sustain these earnings over the long term.
EDIT: From Motley Fool
> Once the head of the pack in computer processors, Intel (NASDAQ: INTC) is coming under serious attack in its core business. Rival big tech companies are opting to develop their own chips, with the latest being mighty Microsoft (NASDAQ: MSFT). On Friday's news that Microsoft is doing the go-it-alone thing, Intel's stock fell by 6.3% on the day.
50% of Server Market is HyperScaler. That is a possibility of nearly half of Server Sales gone. 97% number is no longer accurate as AMD has gain some ( rather small ) market shares in terms of revenue. But Intel is still >90%. ( In Revenue, not Volume )
In the short term most hyperscaler are only going to be switching may be around 50% of their infrastructure, since x86 will still be in demand. But the trend is set and there are no reversal. And Server Segment is the most profitable segment. That is 25% of profit gone in the short term.
And if you now include AMD for the remaining x86 market, you are looking at another reduction. Depending on how AMD is executing ( Which in my book is not nearly good enough on their Server Segment, mainly due to conservative planning for Capacity )
AMD is also for the first time in history really competing in Notebook market with their APU. They have a lot of lessons to learn with regards to working with Notebook manufacture. But again the numbers suggest AMD is making lots of inroad, holding back by capacity again.
You then have majority of the volume driven by OEM and Business usage. These CPU are low margin business. But AMD are not interested in those, and Intel needs those to fill capacity, so they are fine.
So in the next 5 years you are looking at roughly 25% reduction of profits or revenue. Now of course Intel are making their own GPU and are expanding into Network and SSD. And more importantly the market for both PC and Server is still expanding. Which means despite having a lower market share in terms of percentage the revenue number may not be as bad. But again, there nothing even on paper that suggest Intel could change its course. They couldn't even give a valid argument in Investor notes or Technical Forum meetings to clam the market.
Take AMD. AMD doesn’t have the volume or capacity with partners like TSMC to produce at the rate Intel does, even if demand justified it (and I’ll go on record and say there isn’t demand for AMD chips that will match demand for Intel chips across segments and there won’t be for five years, to use your timeline).
Intel is truly on the ropes and is getting screwed on all sides, but AMD is less than 1/10th the size and in this sort of commodity market, size absolutely matters. AMD can’t reliably take the sorts of orders Intel can. It doesn’t have the clout or cash flow to buy the wafers. It doesn’t have its own fabs and is fighting for space and resources against Nvidia and Apple, who are both larger.
It’s great AMD has had such a resurgence (I’m a Mac user but I’m finalizing a build on the first AMD system I’ll have owned since 2003), but they aren’t a real factor in Intel’s decline, outside of adding to the narrative of all aspects of Intel’s business being challenged.
Apple is a threat insofar as it shows the limitations of current Intel hardware, but outside of smartphones, it doesn’t do the volume to pose a real challenge for many of Intel’s markets. Apple isn’t selling into the same places Intel is and across the spectrum of requirements and it is a vertically -integrated company who wants to sell its hardware, software, and services. Apple doesn’t want to sell into the data center (the recent AWS deal proves that very clearly). It’s similar for Amazon. Amazon having its own custom chips for certain workloads doesn’t change the fact that most of AWS’s millions of servers are and will continue to be Intel. Could that change long-term, sure! But it’ll take more than five years for them to get to a point where they can produce those chips in a volume that could match Intel (assuming Amazon even wants to be in the business of making large volumes of its own processors).
Intel is facing massive challenges that will likely be most visible several years from now, and I’ll go as far as to say that the company is wounded right now in a very real way. But economies of scale and R&D cycles necessitate that we’re not going to see a collapse in market in five years.
But TSMC has scaled up quickly before to produce 200m+ iPhone CPUs. And the money is there - Amazon, MS, Google, Nvidia and even AMD are not exactly paupers.
As I've said below Intel's P/E ratio is less than 10. That's extraordinarily low for a company in this market with growing revenue and eps. Clearly the market agrees with me that Intel can't sustain these earnings even over the medium term.
They definitely have the clout, and the volume. TSMC's limited capacity and Apple's ability to pay top dollar is their only real limitation.
As a customer, I tried buying compute sticks in large quantities (i.e. 1000s) from a distributor but they would not extend pricing discounts until Intel approved them. In order to get approved, I had to take online training classes and quizes Intel offered. It was insane. No other company I dealt with required anything like that. I spent hours doing online training on their solutions and never was able to arrange a single discount. Finally said screw it and went with a different solution. Hands down the worst company I dealt with.
Apple designs all its shit.
So does google/facebook/microsoft/amazon
Huawei designs its own OS
China designs/manufactures its own chips too.
It's rumored more companies are designing their own OSIn fact, if you want to prevent anyone else from taking control over your market, you will have to design your own CPU/ecosystem.
Vertical integration at face value is very appealing, it can lead to better products and better prices. On the other hand, Apple, Microsoft, Amazon, are the likes are platform companies. They control the end user's applications, social networking effects, and consumer branding. The OS interfaces, APIs, standards, programming languages that ISVs and a large portions of the software and IT industry rely on. And increasingly leverage over hardware makers, the OEMs, ODMs, silicon designers, chips makers. The first two were getting concentrated first. Think about Microsoft makes Office run only or the best on Windows. Think about Apple is bundling the Apple software ecosystem (Apple Music, iCloud) with the OS and the device. Think about Google bundling its Mapping and office software if you use their search product. But any technology will have a high barrier to entry first and as time passes gets democratized. Chip design and manufacturing is no different. Now it seems like chip design barrier to entry is low enough and platform companies are powerful enough that it makes organic sense for them to dive into the chip design sector.
Fast forward 10 years, it could be that a company will be at competitive disadvantage without its own chips just like how now a company is at a competitive disadvantage without its own ecosystem, platform and multiple products each enforcing each other to create moat. Yelp and Spotify are always at disadvantage compared to Google and Apple Music because they don't have Search and IPhone. This world, likes of Apple, Google, Microsoft, Amazon must have its own chips in order to compete and survive. In that world, what happens to pure play hardware vendors like Intel, AMD, Nvidia? Right now ARM is an instruction set everyone has to agree to because it enables portability, and right now portability is important for survival in the market. But in that world, the market condition will be that portability has disadvantages and "your own thing" has advantages. For example, Apple can add a unique instruction to their CPU that makes a operation 2X faster. Its easily adopted because software developers all use Apple's APIs and compilers so taking advantage of it is automatic. And you want less portability so a ISV can't support your competitor's product as easily. You can imagine this will happen then standards like ARM doesn't matter anymore.
Seems like this tendency towards concentration is natural. My question is at what point does the natural course to concentration stops? Should we artificially put up barrier to stop or slow down this natural tendency before it runs it course (IE, create anti-monopoly regulations, limit a player to enter a different market if they dominate in the other market).
We should always ask: would a project like Linux be possible now? Can somebody become the next Linus Torvalds? If not, then these companies have locked down their devices too far.
Is there an emulation layer for ARM CPU that would allow for near seamless operation of all the golden age games from, say...1995 to 2008?
Having commodity hardware meant anybody could buy a motherboard and parts to build a competitive computer.
What is different now is that the big players try to de-commoditize the hardware.
Result is that you can’t build a competitive hackintosh anymore, as you cannot buy the M1 chip (you can still build a x64 hackintosh, but even that is getting harder, with Macs having various custom chips)
If enough other large companies go the same way, there might eventually not be enough market left to warrant development of competitive commodity parts.
Likewise a Graviton with hardware acceleration specifically addressing AWS tasks can break through existing performance barriers.
The battle of 30 years ago was a battle of ecosystems more than of individual architectures. The market went into consolidation around Wintel and later Linux on the server side.
The market today is orders of magnitude larger and supports multiple ecosystems at different tiers - mobile, laptops, workstation, servers.
Platforms which will not adopt complete vertical integration all the way into their hardware will lag behind.
The other thing is that now, binary compatibility is pretty much preserved and lots of software parts are shared across these platforms. It’s not fragmentation like DEC vs IBM vs Sun vs SGI.
So Apple has iPhone? Now we will have Lumia!
So Apple has iPad? Now we will have Surface!
So Apple has M1? Now we will have something!
Too bad OSX came too late and Microsoft couldn't just copy it by then.
1. A journaling filesystem
2. Protected memory
3. Preemptive multitasking
4. Access control
5. Intel chips
6. Tablet computers
* I say that because Mac OS classic was not the only operating system Apple developed. They had their A/UX operating system from 1988 to 1995 where things like protected memory, preemptive multitasking and the usual stuff from a Unix system of the time were present.
Chipmakers have been crawling forward with progress for years. Meanwhile in house chips like Amazon's and Apple's have been coming out that seem to be miles ahead.
The general theory for this seems to be the chipmakers just don't get how to design their chips to work well with the real software being ran on them. So the mega-corps are just taking matters in to their own hands.