Follow Slashdot stories on Twitter

 


Forgot your password?
Close
typodupeerror
×
Emulation (Games) Windows Businesses Microsoft Software Hardware Technology

Microsoft and Qualcomm Collaborate To Bring Windows 10, x86 Emulation To Snapdragon Processors (anandtech.com) 85

Posted by BeauHD from the two-is-better-than-one dept.
An anonymous reader quotes a report from AnandTech: Today at Microsoft's WinHEC event in Shenzhen, China, the company announced that it's working with Qualcomm to bring the full Windows 10 experience to future devices powered by Snapdragon processors. These new Snapdragon-powered devices should support all things Microsoft, including Microsoft Office, Windows Hello, Windows Pen, and the Edge browser, alongside third-party Universal Windows Platform (UWP) apps and, most interestingly, x86 (32-bit) Win32 apps. They should even be able to play Crysis 2. This announcement fits nicely with Microsoft's "Windows Everywhere" doctrine and should come as no surprise. It's not even the first time we've seen Windows running on ARM processors. Microsoft's failed Windows RT operating system was a modified version of Windows 8 that targeted the ARMv7-A 32-bit architecture. It grew from Microsoft's MinWin effort to make Windows more modular by reorganizing the operating system and cleaning up API dependencies. The major change with today's announcement over Windows RT and UWP is that x86 apps will be able to run on Qualcomm's ARM-based SoCs, along with support for all of the peripherals that are already supported with Windows 10. This alone is a huge change from Windows RT, which would only work with a small subset of peripherals. Microsoft is also focusing on having these devices always connected through cellular, which is something that is not available for many PCs at the moment. Support will be available for eSIM to avoid having to find room in a cramped design to accommodate a physical SIM, and Microsoft is going so far as to call these "cellular PCs" meaning they are expecting broad support for this class of computer, rather than the handful available now with cellular connectivity. The ability to run x86 Win32 apps on ARM will come through emulation, and to demonstrate the performance Microsoft has released a video of an ARM PC running Photoshop.
This discussion has been archived. No new comments can be posted.

Microsoft and Qualcomm Collaborate To Bring Windows 10, x86 Emulation To Snapdragon Processors More

Microsoft and Qualcomm Collaborate To Bring Windows 10, x86 Emulation To Snapdragon Processors

Comments Filter:

  • Will apples one be app store only or will they do same and let it run any app?

  • ARM Server CPUs, x86 on ARM (Score:4, Insightful)

    by HalAtWork ( 926717 ) on Thursday December 08, 2016 @09:23AM (#53445573)

    I feel a transition coming for the low end... Wonder how Intel feels?

    • Intel has seen the writing on the wall for some time now, that's why they've been pushing the Atom and i3 chips to lower and lower power consumption. They need to get some level of parity with the ARM chips or those areas where ARM excels (low power, low heat) will be eating Intel's lunch.

      The Intel powered phones weren't quite there in power consumption, but they certainly had processing power. With ARMs becoming more ubiquitous and general-purpose, you can bet your ass Intel is pushing to keep their har

    • Re:ARM Server CPUs, x86 on ARM (Score:5, Insightful)

      by caseih ( 160668 ) on Thursday December 08, 2016 @02:04PM (#53447179)

      So far in the micro server and embedded space, ARM has been particularly disappointing to me. I have a drawer full of ARM devices I've accumulated over the years. SheevaPlugs, GuruPlugs, RouterStation, etc. All are potentially useful devices, but ARM is hobbled by proprietary boot systems and differing device trees and proprietary supporting hardware. These devices rely on customized linux distributions, and they are often fairly hard to update to new kernels and new flash file systems. Some of these devices have good CPU performance specs, but in practice I've never had them outperform an intel-based server, even a small low-powered one like the atom.

      And now in embedded space we have a plethora of Arm-based devices based on lots of different SoCs from companies all over the world. All with their own forks of Linux. We've got Raspbery Pis, Orange PIs, Pine64s, etc. All very interesting and probably useful. But a nightmare to do anything with in a sustainable way.

      The Pi (and some of these devices) at least is easy to update since everything comes off of the sd card, with no kernel flashing required. And some of them like the Pi have a fair amount of hacker inertia behind them, so they are capable of doing cool things (maybe not as server replacements though).

      With x86-based embedded systems and small servers, at least I can run more standard, off-the-shelf distros on them. I'd far rather deal with a conventional linux server than a sheevaplug, even if the sheevaplug is a nice tiny thing with lots of potential.

      In fact my current home office router is a small, low-power Intel-based computer running bog standard, minimal install of CentOS 7. Wifi is hung off of that using a consumer-grade access point running in bridge mode.

      If arm devices had a standard boot process like ufi or even the bios, and could boot off of a variety of devices in a standardized way, including ssds, hard drives, usb sticks, and internal flash storage, and could run stock distributions downloaded from distribution web sites, without custom kernels, then I'd say for sure x86's days are numbered. Arm is good at remaining fragmented though.

      • Re: (Score:1)

        by Anonymous Coward

        The ARM platform mess is likely one of the reasons behind the rapidly growing support for the RISC-V [riscv.org] architecture. The more obvious one being that no one wants to pay rent in perpetuity. ARM may not be as bad as intel, but they are definitely still a substantial burden on hardware developers, and a good open architecture and platform is what just about everyone wants. With Android being mostly platform neutral, I think ARM will also be in for a rude surprise, and turn out to have been a foolish investmen

    • I wonder if they'll transition Windows Server as well https://hardware.slashdot.org/... [slashdot.org]

  • Performance? (Score:4, Informative)

    by crunchy_one ( 1047426 ) on Thursday December 08, 2016 @09:51AM (#53445683)
    According to TLA, x86 compatibility is achieved through emulation. Emulating the x86 instruction set is a non-trivial exercise that almost invariably results in extremely disappointing performance. Why? The x86 instruction set is an accretion of the instruction sets of older Intel processors, beginning with the 8008. This yields a difficult (i.e., computationally expensive) instruction set to decode and execute. Over the years, Intel has implemented micro-architectures that address this problem through special purpose hardware. If you're so inclined, have a read here http://www.intel.com/content/w... [intel.com] for details. The takeaway is that simply emulating the x86 instruction set results in about a 100x slowdown for an equivalent clock rate. So, although this is an interesting technology demonstration, I seriously doubt it will prove useful outside of a small set of applications. It will certainly not be a satisfactory gaming platform.
    • So you're saying it'll run Crysis?
    • "The takeaway is that simply emulating the x86 instruction set results in about a 100x slowdown for an equivalent clock rate. So, although this is an interesting technology demonstration, I seriously doubt it will prove useful outside of a small set of applications. It will certainly not be a satisfactory gaming platform."

      Sounds like you didn't read the article or watch the embedded video, where MS show x86 Photoshop and x86 World of Tanks being emulated on Windows 10 ARM and a Snapdragon (835?) SoC - a
      • Don't believe everything you read/see in a press release. Apply some critical thinking.

        • Re: (Score:2)

          by ET3D ( 1169851 )

          I believe press releases (particularly when coupled with demos) a lot more than I believe random people on the internet. That's applying critical thinking.

        • Don't believe everything you read/see in a press release. Apply some critical thinking.

          A reasonable person that both watches the video and reads your comment would conclude that either you are mistaken, or Microsoft and Qualcomm have somewhat overcome or mitigated the issues you point out.

    • Re:Performance? (Score:5, Interesting)

      by Halo1 ( 136547 ) on Thursday December 08, 2016 @10:26AM (#53445885)

      The takeaway is that simply emulating the x86 instruction set results in about a 100x slowdown for an equivalent clock rate.

      Emulation definitely results in slowdowns, but it's generally much less than 100x. In particular since any emulator that focuses even slightly on performance uses dynamic compilation: it translates the code once from x86 to the host architecture and from then on runs this translation. The translated version will probably be less efficient than the original code, but by no means 100x slower. 2x to 5x seems more realistic on average, although there are certainly outliers (e.g. code that intensively mucks with system registers or that triggers context switches will be slower, while some straightforward calculation loops may actually become just as fast as or even faster than the original code depending on the target architecture's nature).

      • I believe that Dynamo/RIO or how that thing was called actually managed to occasionally produce code *faster* that the original, if said code was generated with an inferior compiler.

        • Re: (Score:2)

          by Halo1 ( 136547 )

          Back then it was still Dynamo. And they only managed to do that on a particular HP PA-RISC architecture, because it was very sensitive to instruction cache missers (or had a bad branch predictor?) so that creating linear traces of code was very performant. They later tried to reproduce it on x86 and failed horribly (just like I did during my master's thesis; the best I got was a 20% slowdown for gzip, I think the best they got was no performance loss with some benchmarks).

          • If you could get (say, on ARM) that 20% slowdown for code transpiled from x86 binaries (with a native build being the baseline), I guess you'd be very happy? And so would we. This *could* be useful.

      • In particular since any emulator that focuses even slightly on performance uses dynamic compilation

        Except a lot of mass-market ARM platforms nowadays have W^X security policies that ban dynamic compilation.

        • Re: (Score:2)

          by Halo1 ( 136547 )

          That only means you have to mark the pages containing the code you just generated read-only once you're done.

          • That only means you have to mark the pages containing the code you just generated read-only once you're done.

            Several operating systems in wide use, such as Apple iOS and the operating systems of modern video game consoles, offer no way for third-party applications to switch a page from read-write to read-execute. When a page is allocated for data, the OS clears it first, and it stays non-executable until deallocated. Only the OS's executable loader* has the privilege to allocate pages for code, and once the loader loads a module, verifies its digital signature, and flips its pages from read-write to read-execute,

            • Re: (Score:2)

              by Halo1 ( 136547 )

              Given that this is a Microsoft-sanctioned emulator that they will include with their own OS, I think you can be pretty confident it will have similar permissions.

      • Much less than 100 times alright, but somehow most JITs/emulators (hello Transmeta) manage to incur a ridiculously high load on the system so running multiple large emulated apps simultaneously is troublesome. Why not distribute in CIL or llvm IR format?

        • Re: (Score:2)

          by Halo1 ( 136547 )

          Having an intermediate format that you statically translate into the target architecture is definitely useful (like Android is now doing with ART), but keep in mind that LLVM IR is not architecture-independent most of the time. E.g., when LLVM IR is generated from C, then this C code will at least have been compiled based on a certain pointer size, size of long, size of long long, alignments for struct fields, etc. CIL is better in this regard.

          However, you should see this as a solution that will be used in

    • Re: (Score:2)

      by GoRK ( 10018 )

      Your information is highly dated and perhaps your sources are also a bit biased. At any rate, 100x performance hit is stupid wrong.

      Static translation was achieving 50-70% native performance rates (measured against clock cycles) with FX!32 on Windows NT for Alpha in the mid 90's. The problem of course has been very well studied since then particularly with the advent of virtualization and the x64 instruction set and the need to enhance the performance of x86 code running on even Intel's own platforms. Furthe

      • The applications that lose in this scenario are the ones the rely on raw single thread performance. Certainly some games are in this camp, but many games which make efficient use of threads are not.

        It's not that simple. Games are multithreaded now, yes, but they do not have a crapload of threads which can make use of a crapload of cores. If you're taking a substantial clock rate hit and another substantial hit from translation overhead, the truth is that it's not just the high-end games which are going to suffer, nor the low-end ones, but any of them which are not very old — as defined by coming from the era when PCs had even lower clock rates. It's already true that Intel processors with less c

    • It depends on if they emulate it by translation and shadowing, or by interpretation. Software translation is rather fast, but not native-fast. To get native-fast, you have to go native.

      I've been suggesting an accelerator chip (maybe even off-die) that decodes x86 instructions into the internal RISC instructions stored in the ICache, but people keep telling me it's impossible because... they're stupid. Modern x86, x86-64, and ARM chips all read instructions in their ISA and translate to an internal CPU

    • I think it's safe to say it's slower, but as I used to remind the engineers when they pushed back against releasing slow solutions when I was working as a liason between customers and engineering: "not working is infinitely slow".

      It's easy to see how Microsoft wants this for the Surface Phone since it will allow customers to run x86 applications on their phone. If your choice is between not running an application and running it slowly, most people will always choose "slowly". Most Win32 software isn't h

  • Not all emulation (Score:2, Informative)

    by Anonymous Coward
    Only the Win32 x86 apps are emulated. The Win64 bit ones are not, and neither is the OS. That is why the performance is good (check the video). Hope that helps.

    • Re: (Score:2)

      by Guspaz ( 556486 )

      The Win64 ones aren't emulated because they simply aren't supported.

      • Re: (Score:2)

        by PCM2 ( 4486 )

        That's not actually that big of a downside. With Microsoft Office, for example, Microsoft still recommends most users install the 32-bit version, even though almost everybody is running a 64-bit OS these days. The exception is people who need to run crazy big Access databases (or ... shudder ... Excel spreadsheets).

  • FX!32 (Score:2)

    by DrXym ( 126579 )
    Microsoft produced a version of Windows NT for DEC Alpha and threw in x86 emulation layer called FX!32 so it could run existing software. It worked but it ran like a dog, far slower than native x86 instructions. And not for want of trying because it did machine translate instructions to try and execute code natively.

    I really don't expect the picture to be any different with x86 over ARM. I expect they'll machine translate x86 instructions into native ARM instructions in some way and cache them somewhere,

  • Apparently Microsoft hasn't gotten the memo about restrictions not being wanted, so they keep on trying to do it?

Slashdot Top Deals

Math is like love -- a simple idea but it can get complicated. -- R. Drabek

Close