Microsoft already have a developer kit called Project Volterra for making Windows apps for ARM so they're definitely serious about making their own ARM chips.
i mean, arm has existed in the microsoft ecosystem for a long time. surface RT and windows phone are the consumer examples, but you could still buy versions of windows embedded for arm, mips, etc.
@@pigalex Yeah but would you considering how bad they were? Volterra's goal is to (hopefully) make Windows on ARM not suck like how it did the last few times.
They are not serious. The x86 to ARM emulation is hilariously bad. Rosetta on Apple M chips just works. ARM Microsoft Surface Pro laptops can't run 99% of x86 software. Most stuff just crashes.
Given the power consumption in MSFT Data centers (The majority of it runs on Intel cpus), I suspect this development would be more of a data centre centric move in the first few iterations of any architecture they come up with. Dispersing all the heat generated by large data centers can cost as much, if not more, than the energy used to power the servers. So a more power efficient cpu, that runs cooler, is an obvious target to aim for. Any savings made via power efficiency would, of course, be an internal profit used for greater expansion and capacity of it's existing data center estate. Just my 2 cents.
We're also at the physical limits of transistor size, without Moore's law to reliably increase performance we're going to have to see efficiency and performance gains from other methods (like switching to ARM, packaging improvements like what is already being done in Apple Silicon, and of course optimization of software itself... maybe rebuilding Windows from the ground up to run on ARM is the way forward no matter how painful it is).
Microsoft’s cloud predominantly runs Linux, not Windows. So yes, moving away from x86 there would be more feasible than trying to do it in the retail market.
if we're going for power efficiency, wouldn't it be easier to go for epyc? it doesn't need making architecture by yourself, polishing it with 2-3 iterations and rewriting whole code used on x86, at this point swapping intel to amd would be most cost effective in short run and then they could take their time and short term earnings to be put into designing i have no idea what cooling costs are in this case and how much relatively it would save them per month basis, but after x time it should be net positive "AMD EPYC offers up to four times greater performance per watt than Intel Xeon, meaning that users need substantially less electricity for operations, resulting in significantly lower operating costs (TCO)" ~exittechnologies 4x better efficiency almost means 4x less heat to get rid of for same workload, it's probably not that linear, but halving this cost would make a difference in a long run and probably before they can get working x64 systems
@@lawrencedoliveiro9104 Except most of Azure's workload runs on x86, they've slowly begun introducing an Azure ARM platform but the vast majority of their customers run x86 code on Azure. The AAD and webhosting customers might shift over to ARM servers fairly easily, but all the customers running Linux/Windows workloads on Azure can't really be shifted over.
Without a translation layer like Roseta2 on macOS, to emulate x86 on ARM, or just straight adding x86 hardware translation in to the chip it will be a failure. Apple users are more or less used to apple ditching support after a couple of years for their software, on a Windows I just can run apps that run 20 years ago, mostly without bigger problems, and if they do hard cut off support to move to ARM it might bite them in the ass, because they make Windows users mad.
This is where I think Windows needs to be split. Distribut the OS in a 'light' configuration without all those legacy drivers and oddities. Then provide an optional service pack to add it all back in. The vast majority of Windows users dont need a driver for a 1997 dot matrix printer and it's incredibly moronic of Windows to still be including all that stuff in its mainline build.
y'know there's windows for arm cpus, right? and that it already has x86 and x64 bit emulation built in? that way for example some guys ported drivers and emulated uefi on phones like OnePlus 6t so you can install and flawlessly run Windows
@@Insky_ I should be more precise, Windows on ARM should have this emulation done good, on a level like in Roseta 2, because as far as I know, for now this emulation is terrible, and it is not important that this is a fault of the chip, or the system, it is just bad and have to be way better.
@@Rick-vm8bl idea maybe great, but hard to achieve it probably, maybe even harder than writing new windows from scratch and add some optional "legacy layer" on top. And we are talking about years of development and cost around 1-2B dollars, probably, to write a new OS from ground up.
@@slizgi86Apple benefits from doing making specific changes to the hardware that makes X86 emulation easier. It’s not efficient purely because of software so unless Microsoft also develops their own chips, windows on arm is not gonna have efficient X86 emulation.
Microsoft also has incentive to do so to build up their tooling for their cloud business. AWS and Google already make custom chips for a lot of processing and cloud workloads.
Yeah, I doubt M$ is trying to make Windows on ARM again, even they aren't that dumb. They are probably trying to compete with Graviton and Ampere Altra in Azure.
To think that the surface lineup, which when it launched almost 10 years ago wasn't exactly seen as a good laptop system or tablet system- is now somehow still going successfully and being used as the center of windows for ARM
Keep in mind that Amazons entire product line is based on a very narrow codebase and ARM licensing. Also keep in mind that many many user experiences (aka apps) are now Cloud and Web centric. Wintel was for desktop and server - ARM fits nicely into Internet appliances. Although one could argue x86 SOCs are decent - just higher priced. Back to servers - the big mama jamma out of Alibaba is 128 cores / 128 PCI lanes ARM based. Scales well with a lower power consumption. That matters in massive DCs.
It's simple: the chip market went from "I have to buy chips that a few companies design and manufacture themselves" to "I can design my own chips and delegate their manufacture to a third party". Designing chips is complex, but nowhere near as complex as manufacturing them, so several companies (Apple, Nvidia, AMD, etc.) are taking advantage of the fact that a company called TSMC has the technology to manufacture the most advanced chips in the world, and the companies now only have to design the chips.
the reason why microsoft is doing is is because companies want to charge licencing fees to use those Machine Learning accelerators, and the prices are insane, look into how much Qualcomm's XR2 accelaterator license costs
@@SplitScreamOFFICIAL Thats not much, lol. Compare it to the cost of building that IP in-house; the number of employees earning six-digit salaries working year(s) on the project... that's nothing at these scales.
@@jrcowboy1099 it is when you're a game dev company that also wants to use those because Facebook can't distribute that license downstream, $100,000 is enough to make an indie company lives or dies
Honestly it sounds like the opposite now days with the competition between x86 and ARM, since it's hard to get all the developers to develope for your platform
I've been expecting Microsoft to bring out a full blown high performance RISC-V CPU, complete with a large Vector Processing Unit instead of fixed sized SSE / AVX SIMD.. This scales much better as you can have lots of simple cores and a separate VPU. Integrated Graphics bought off ARM for not a lot on top. They have put quite a bit of money into RISC-V as a concept. It makes more sense than going fully ARM, although starting off with ARM may be a quicker way to get their foot in the CPU market door.
If they make it a success, I wouldn't be surprise that they would slow other CPUs (or even lock them out). Look at how persistent they are on their IE and EDGE
this is why i decided to daily drive linux mint since the start of this year. anti-competitive, depredatody, monopolistic behaviour. my mental health is so much better. Haven't had the need to touch windows since then.
@@GuyGamer1Bruh, why are you lying on the internet? Edge is based on Chromium browser. The AdBlock extension you are using is a Chrome extension. The moment Google disables manifest and prevents AdBlock from working, all Chromium Browsers will STOP working with the AdBlock unless the AdBlock devs find a different way. ALL of THEM. They all use Chromium FULL CODE, not just the engine.
efficiency comes mostly from the best available process and reasonable clocks, AMD is there and Intel isn't far off when undervolted and run with conservative power limits 7950X capped at 90W basically matches the M2 Ultra R23 score, for me that works pretty well, doesn't it? M2U probably does much better in memory intensive tasks, but that also comes with a huge cost
That is not true, in a CPU every transistor is using energy regardless if its being used or not. So a CISC CPU like x86 that needs 10 times more transistors to handle 10 times more commands will be using 10 times more energy than an RISC CPU like ARM regardless if 90% of those extra commands are almost never used any more.
@@hubertnnn This is objectively false with CMOS transistors, the only ones used since, what, the 70s? In fact, this is a huge reason why AMD's new GPUs are becoming more compelling. They even state they are doing exactly this in their announcement for their most recent GPUs. Essentially, this is what 'clock gating' means (oversimplified ofc).
@@jrcowboy1099 also how A2000 has the 3050 performance at less than half the power using the same archi, and 4000 Ada is probably around 4060Ti in terms of performance again with half the power the more you push clocks the more voltage you need and the more*more power you waste, for CPU that makes some sense due to poor scaling of many tasks, but GPU scales nearly linearly with "cores" count (up to some point, but 4090 still works well in that aspect) and lower clocks allow higher density = higher yield than the "regular big die"
This seems like the response for the Chip Shortage of yesteryear. This timeline would make since. It happened, a year or 2 later they do something about it.
Windows uses a Hardware Abstraction Layer (HAL) and provided 3rd party apps aren't coded to closely to the hardware then a move to ARM architecture shouldn't be much of a problem. As for x86 though, it carries a lot of historical baggage and really we should be taking advantage of the CPU architecture progress of the last 40 years and move to something new. But be warned that doing so won't be painless.
I'm still amazed at how much intel was able to improve x86, I thought it was basically as good as it was going to get around nehalem. But, I stll think the fundamental limitations of x86 will lead to its demise eventually, and maybe this is the beginning of the end, or maybe intel will manage to stretch it out a little longer with this upcoming deprecation of old features like real and protected mode.
x86 at this point is no limitation to performance, the only advantage of arm is decoding complexity, but today's chips are so large and complex that decoding complexity is not a problem at all, intel is removing those old modes for increased security not performance.
@@odnx x86 ISA is CISC, but x86 CPUs have switched to RISC cores + CISC Decoders over a decade ago. CISC is easier to program for, at least in assembler, but RISC has better performance and efficiency.
When talking about 32b & 64b processors it actually applies to many functions/busses. The answer is complicated. * Instruction size or base unit; * size of integer datatype that can be processed by ALU per instruction; * Data widths to RAM. For current CPU's: **Some 64b CPU's already allow 64b CISC instructions to process larger numbers or SIMD just by using shorter operand codes added to the basic 64b instructions; **There is no urgent need for >64b of RAM address space; ** data buses can also be wider than 64b. Some modern can handle 512b of data in 1 instruction because they process several small values burried within the large larger, 80b FP, 128b values.
What i want is something like AMD already does. ARM cores on an X86-AMD64 APU But instead of a sub 1W single core(i think) ARM core just for OOMB like DASH, make a higher level ARM segment, say 6 A53 cores, that Windows and some store apps can run on at under 10w, shutting off the rest of the chip, until you need X86-AMD64 compatibility.
The problem with windows on ARM is not windows, it is EVERYTHING else. No business will switch to an ARM only chip, when many businesses are using softwares that are over 20 years old, from before we had AMD 64 bit. Businesses would be far more likely to switch if it had ARM+x86+AMD64
The thing, in my opinion, that has been the *most* impressive about Apple's M1 and M2 chips, is the Rosetta 2 software that lets you run Intel-based Mac software on their Apple Silicon chips. That has been nearly flawless, and has been a key component of the success of Apple's custom silicon. Microsoft's "Rosetta 2" equivalent will make or break their custom silicon for the average consumer. If it seamlessly runs existing x86-based apps, people will have no problems buying Microsoft's custom-silicon devices. In the "Windows on ARM" arena, btw, I've been running Windows 11 Pro ARM-edition in Parallels Desktop on my M2 Pro MacBook Pro for about 4 months now, with no issues whatsoever -- but, I haven't been stressing it badly, either. It's not like I'm playing AAA games on it or anything (I have an x86 desktop PC for that). I just use it for Microsoft SQL Server Studio on my Mac's Win 11 installation, because that particular piece of software simply won't run on a Mac. But it has been 100% flawless for that task, so, maybe there's hope for Windows on ARM.
Revival of Windows Mobile? They worked hard on touch interfaces, which would be found in Tablets and other more mobile devices. It doesn't make a ton of sense to invest so much into Touch if it's only for those enthusiasts with 2-in-1s. They do have a MSVC for ARM compiler, so *technically* everything would be ARM-compatible if you, well, re-compile it. The problem is ARM-specific code won't run on x86 computers, so a side must be picked regardless.
It won't. Windows Vista didn't do it, Windows 8 didn't do, Windows 11.... Somethings are too embedded to shift. The only win I see for Linux is the Steamdeck. Chromebooks failed.
Actually, Linux owns a huge swath of the server environments. It has never been able to break into desktop because Windows was agnostic on the hardware. If Windows try to mimic Apple everyone will either migrate to Apple, why not go with the company that knows how to do the walled garden best, or Linux because it will be the last multiplatform left. I will never understand why Microsoft, a company that once owned around 97% or the desktop market, decided to follow Apple, a company that owns less than 10% of the same market. Maybe the Balmer and the new CEO just hate success with the end user.
I wish more gaming laptops have desktop cpus like the origin pc eon-17x since it has the Intel i9-11900k desktop cpu. The apple m1 and apple m2 is manufactured by tsmc like the tensor on the pixel devices are manufactured by samsung. Samsung and huawei manufacturers there own in house processors. I wish the new exynos processors were like the samsung galaxy note 5 since all samsung galaxy note 5 was using the exynos processors. The samsung galaxy note 5 was the last note series of phones that was using micro usb before samsung switched to usb c starting with the samsung galaxy note 7.
1. Windows NT was built for non-x86 architectures from the start. NT 4.0 was available on four different architectures, XP on three (x86, x86_64, and Itanium). 2. The modern-day Windows 11 on ARM has emulation for x86 software, and so does macOS. Unlike macOS, Windows prides itself on backwards compatibility, and a lot of people and businesses depend on ancient software that solves a problem and is good enough. Apple regularly kills off older apps (most recently, they axed 32-bit apps a year before introducing Apple SIlicon). Microsoft knows this and will not leave old software behind.
Don't forget trust issues. This is another non-technical obstacle Microsoft will need to clear. How easy will it be to get Linux running on it? Will it have baked in telemetry? These days it's not just the technical potential of making their own CPUs that a company has to contend with.
Windows needs to be totaly rewritten, and anything related to older compatibility should use virtualization tech (like Apple does), and not ship the same os every year with a coat of paint on the outside
With how impressive Apple's latest architecture is I wonder if ARM will be the future in general. Even if it is, Microsoft transitioning Windows to ARM might be a difficult proposition for end users given that most Windows programs are still made with x86 in mind.
Unless you can game on ARM it's doomed to stay niche/mobile. Just like Linux. Doesn't matter if it's better. We don't want to do 2 things well, we want to do all 200 things, even if they're not efficient and safe. They're still cheap enough to be worth it since there's so many.
@@estiennetaylor1260 ARM is not a mobile, it's a simpler architecture by design doesn't process complex instructions, and believe it or not most apps and games don't require the complex instructions of x86 and can run just fine on ARM, only server stuffs and some very niche subjects require the complexity of x86... I mean take a look at ARM apps and games on Apple silicon, they run just fine, even Microsoft was among the firsts to release a native ARM port of their entire Office apps, then Adobe followed then games engines from Unity to Unreal then all kinds of software are being ported and runs without issues, proving you don't need the complexity of x86...
I think yall are wrong about the microsoft sq series. Am writing this on a surface pro X (sq1) and its great. Basically everything from microsoft runs natively and even emulated software just works. You just install any exe and it WILL work if you got enough power. (for anyone thinking about one, i bought used and i FULLY Reccomend going for it if you want a universal machine, from note taking through light gaming or photo editing. Also its so goddamm thin and LIGHT! btw linus, if your reading this, take a look at it again, its WAAYYYY better than it was on release, it was just ahead of its time.
They look nice, but build quality is not that great, their failure rate is far too high. If yours hasn't had any issues I'm happy for you, but unfortunately they have built up quite a bad reputation after years and years of failures. Also while x86 software does run on it, plenty of software can't as it isn't fully compatible. Basic software (Chrome browser and whatsapp for desktop or something like that) will work, anything more complex usually crashes. Performance is abysmal, it's good performance compared to a tablet, bad performance compared to a laptop and terrible performance compared to x86 desktops. Glad they're working out for you, but at the same time there's a good reason they have a bad reputation...
@@someguy4915 You are wrong. Chrome and whatsapp works great (am using firefox They have a bad rep, because of early reviews while the software wasnt finnished yet. They also arent built bad, maybe some first batch or something, but mine is built great!
Windows was with Pocket PC (PDA) back in 2000, which was a arm BY THE TIME THEY'RE GOING TO PUT THE BRAIN TOGETHER,I WOULD HAVE MADE IT MYSELF ALREADY. (WIN-CE)
The whole advantage (for the consumers) of the x86 is the fact it is tied to the IBM PC architecture. ARM don't have such "restrictions" and could easily turn into the same horrible thing that is android, where you're forced to use the vendor version of the OS and all the unremovable garbage that comes with it. Imagine having hardware locked mcafee anti virus.
Well, with a bit of fancy softwares and some good luck you can install every Android version you like on your phone. Deleting pre-installed apps is quite easy, you only need a PC, internet and a cable to connect your phone to the PC 😁
@@lucadominguez4659 This depends on your phone model. Yes, I may be typing this to you from GrapheneOS, but I specifically bought a phone with an unlocked bootloader so that I could install it. If your phone doesn't have an unlocked bootloader, you can't install other Android roms, and many brands of phones do not have unlocked bootloaders. (It may still technically be possible to install other operating systems, even with a locked bootloader, but by that point we're likely talking hardware modding your phone. Most people, even technical ones, aren't going to do that.)
The argument that m1 chip is more battery efficient than their x86 counterparts needs to be looked at better. We are getting 5nm x86 chips on laptop soon. If their battery life is similar to apple then I don't think that argument holds water. Plus X86 is going to become more efficient once they get rid of 16 and 32 bit instructions.
pretty sure the lattest amd apus are already as efficient as apple's and as soon as they also move to a unified memory architecture for bigger apus then there will be no advantage for apple except the node they're in
it's been part of Microsoft strategy since day one. this is why we had things like the Microsoft mouse and keyboard, and later the Microsoft game machines, they are positioning themselves ... :)
Microsoft actually has a reason to do it, and I wouldn't be surprised if they have their own chips. Microsoft has one of the largest cloud systems in the world. This means custom silicon to handle certain data center loads. For example, RU-vid has its own chips designed to process video. Microsoft has done it in the past. They got IBM and AMD to design parts of a processor, and Microsoft paid them royalties. This was for the Xbox 360.
Microsoft will push for its own chips, then all the market players will push for linux operating system which will be customizable for the use cases for each organization, that's just my prediction.
Ages ago, with a 386 Computer, I wanted to change to Windows. I bought Win 3.1 and suddenly realized NONE of my software would work on it. It must have taken 3-5 YEARS for these random, screwball companies to catch up by re-writing all of their programs. SO, I wonder how long it will take this time? I would think much faster! ANY IDEAS people?
What if you make a arm based window and just add 1 x86 core that you can stop or start as needed for compatibility ? Like settings, battery life, enable x86 support. Can you have a system that throws a program to a core like that ?
3:26 not actually true, windows nt was cross-architecture from day one. hell, windows was even reported to arm twice; both with windows phone and the surface rt.
The entire server industry has been trying to get ARM to equal x86 in performance and so far the only ARM chips that get close burn more energy... ARM is just not good at emulating x86 quickly, at least not without giving up ARM's main benefit: energy savings. This whole 'ARM will disrupt the x86 market' has been getting yelled around for over 25 years now and so far the only thing ARM has shown is that it strangely encounters the same issues x86 does: the laws of physics... Parts of Azure already use ARM servers; for ARM workloads where customers specifically choose an ARM platform. Replacing the x86 part of Azure (the majority of Azure) is not going to happen within the next 10 years, if ever.
If you look at the list of the world’s most powerful supercomputers, you will see that half of the machines run non-x86 architectures. And all of them run Linux.
hope it isn't a failure, cause then I can make the comparison of Microsoft making CPU processors the equivalent of Intel making video cards... some companies need to stick to what they are good to and not dive into products they are not good at (like Intel with their joke video cards).
I think Intel's new x86-S is really onto something... a simpler, newer instruction set that gets rid of the backwards compatibility and allows for growth.
For its powerful capability its should be personal so no need integrate with network no wifi no blue tooth , no speaker no mouse only projector laser keyboard and musics keyboard and laser projector display is for personal
While ARM chips have done great things for Apple, it's also not a silver bullet. Doing a pure SoC design has its drawbacks and they are ENORMOUS when the rubber meets the road. Great performance per watt, but definitely not the best performance when efficiency is no longer a priority. The thermal ceiling of the chips is so much lower than x86, but Apple proved that it's still possible to thermally throttle them and it's hard to avoid when you have the entire damn computer on one die. If there's a way to replace x86 chips with RISC architecture while still allowing people to mix and match the other single-purpose components like GPUs and RAM, I'm all for it. But we all know that won't be easy. They should force all of the top OS and app developers into a room and lock the door until they decide on one open, universal, reduced-instruction language to follow. Make it into a consortium, like JEDEC or something.
It would make sense for Microsoft to go down the path of RISC-V machines. There wouldn't be royalties to pay and since the core is a lot smaller, Intel could make more cores per CPU. Really it should be a 64/128 bit version of the RISC-V. Nearly every machine is moving to 64 bits today. A 128 bit bus in selective areas could help them get effectively more IPC.
I'd imagine they're building for azure though, not PCs? In that environment the software which is running is far more controlled and Microsoft is directly paying the power bills.
SQ1 wasn't microsoft "custom" it was designed by Qualcomm. Word on the street is that the next gen Qualcomm Chips (designed by former Apple engineers acquired when QCOM bought Nuvia) are going to be pretty sick. search Qualcomm Hamoa.
There's a reason why x86 has stuck around so long. Demand-Side Economies of Scale and Backward Compatibility are a vast, self-sustaining behemoth. Most people aren't willing to spend what little money they have on a new standard that will likely fail. It's a bit like trying to convince everyone to ditch the languages they spent a lifetime learning for Esperanto or another 'better' language.