more on ARM vs Intel

Avatar
13

First, read Paul’s “This is When Windows 10 on ARM Will Finally Make Sense” if you haven’t already. Qualcomm’s 2019 offerings will be significantly and noticeably better than Intel’s, particularly in tablets. I will be shocked if the Surface Go 2 doesn’t use a Qualcomm chip. Here’s why: Intels 10nm situation isn’t the only thing hurting it. Because of the way Intel works, they can’t move to their next microarchitecture (Ice Lake) without 10nm, so we are stuck with the now vintage Skylake microarchitecture (they just keep renaming it every year, kaby lake, coffee lake, whiskey lake etc.). That’s a double whammy. And to make matters even worse, this platform doesn’t support modern LPDDR4 RAM, which is very important in mobile devices. 

So next year, the Surface team will have a choice between the successor to the Pentium Gold, which will not be significantly or noticeably better in any way (I’m not even sure Intel is making a successor to the Pentium Gold!) or an upclocked Snapdragon 855, which will be made on TSMC’s 7nm process (which is significantly better than Intel’s 14nm++ process in power efficiency and heat output). Also, the Snapdragon 855 will use “LPDDR4x” RAM, which is even more power efficient than LPDDR4 RAM. The difference between the Snapdragon 855 platform and Pentium Gold will be almost shocking. I hope Intel can get 10nm figured out ASAP but we’ll just have to see…

Comments (13)

13 responses to “more on ARM vs Intel”

  1. Avatar

    Jhambi

    Isn't Apple also going to dump intel for their in house A chips? Goodbye x86

    • Avatar

      curtisspendlove

      In reply to Jhambi:

      I don’t see them totally dumping Intel. They *could* but there are a lot of the “pro” market that need “x86” architecture for at least another 5-10 years. The computer industry is slowly moving to ARM as it makes inroads in the server community and elsewhere. But that will be a slow march.


      Going full “A” in the consumer lines would not surprise me, however. (Most people don’t have to cringe at the thought of emulating Windows on ARM under an “A” series emulated Boot Camp. Though I would be surprised if Apple spent resources on even allowing things like Boot Camp to run in an “A” series Mac.)


      And it would be an additional differentiating factor for the higher prices of the “pro” lines.


      (These are the thoughts that keep a Linux PC build in my back-pocket in case Apple ever does yet another crazy thing.)

      • Avatar

        skane2600

        In reply to curtisspendlove:

        Just to point out for what you probably already know, switching a Linux server running on Intel with one running on ARM is pretty simple compared to the same transition on the desktop. I'm not convinced that the server switch is predictive of what might happen on desktop systems.

        • Avatar

          curtisspendlove

          In reply to skane2600:

          Correct. Sorry I wasn’t very specific there. I simply meant until (unless) the server switch to ARM happens for all portions of common server workloads there will always be niche needs for Intel x86-64 desktop hardware.


          Although to be fair, containerization is trending toward making that a much less risky (scary) proposition for tech teams.


          Still, one generally prefers to run the same stacks and architectures across development, staging, and production.


          It isn’t absolutely neccesary, but it is comforting. And when the consequence for getting it wrong is working outside regular working hours, it gives added incentive. ;)

          • Avatar

            wright_is

            In reply to curtisspendlove:

            ARM is also, currently, an unproven (and generally unsupported) platform for hypervisors.

            There are a few hypervisors around for ARM, but the biggest are KVM and Xen, which are still relatively thin on the ground in on-prem.

            That isn't to say that it won't happen, but most companies are not in a position currently to switch - SPARC and PowerPC seem to have more hypervisor solutions available than ARM, currently.

            I read a study on the VMWare site, ARM has some advantages over x64 in some places, but cache and prediction seem more open than on Intel, and we have seen where Intel is with Spectre and cache sidechannel attacks. The ARM designs are coming out of a "simple" mobile world, where the architecture is "simple" in comparison to the data center. I'm not an expert, but it looks like ARM still needs to undergo some of the design changes that x64 has already been through, before it is really ready for data center use with hypervisors.

            On the positive side, ARM have the advantage of Intel's hindsight, so hopefully they won't make the same mistakes going forward.

  2. Avatar

    Paul Thurrott

    Yep. Next year could be very interesting.

  3. Avatar

    longhorn

    I'm not saying Intel doesn't have problems, because they do. Here's the thing; Windows Phone failed, not because it was bad, but because of lack of apps. Windows on ARM is in the same situation. There is hardly any software for Windows on ARM. Lackluster Store apps and emulated x86 apps. It doesn't matter if ARM is technically better. Intel has the apps (x86 and x64 at native speed). ARM against Intel is like Windows Phone going up against Android and we all know the result.


    I have nothing against ARM but I don't think they stand more than a snowball's chance in hell. MacOS and ARM won't happen for the foreseeable future for the same reason. Apps are what make or break a platform. If your favorite browser, game and business program don't run well, why bother?


    • Avatar

      curtisspendlove

      In reply to longhorn:

      MacOS and ARM won't happen for the foreseeable future for the same reason. 


      I’m not so sure. I can’t shake the feeling that the 12” MacBook was (ultimately) suppose to run an A-series chip (at least a quick revision to 10nm “i” equivalents) and it just wasn’t ready yet. I think Apple has been hard at work on this for a while.


      And I think we’ve only heard half the story for the “sneak peek” IOS on macOS app story. I think “end-game” for iOS apps on macOS is for them to run natively. (But I think they’ve hit plenty of speed bumps with macOS on ARM. Another reason I’m light on criticism toward MS for Windows on ARM. This stuff is hard, folks.)


      (I believe the original MacBook 12” was supposed to release in 2015 with the 14nm “m” Intel chips, be revised in 2017 with 10nm “i” chips, then move to Apple “A” SoC around 2020. I think there are plenty of hints for that if you look back nd forward in Apple’s timeline. I find it curiously suspicious that it has not gotten measurably superior or cheaper since its release. And I also find it curiously odd that the MacBook Air has sat, mostly stagnant, for a very long time for such a good-selling laptop. Apple is always puffy and self-congratulatory when releasing new products. But they were exceptionally puffy when announcing that MacBook in 2015. They went to *great* lengths, even for Apple, to make it all super svelte.)


      I do agree that it is the software primarily holding ARM back. And it isn’t quite as easy as just recompiling everything for ARM. But there is a lot of tooling that can help.


      But I do wonder what “foreseeable future” will be. I think we are going to see this stuff come into its own near 2020 and in the early 20’s. (Heh! It’s going to be fun referring to the 20’s.)


      I definitely consider 2020-2025 to be “foreseeable future”.


      Intel stuff is certainly not going away though. And I think they are going to be significantly more powerful when they get the process shrink (it will finally drop the thermal envelope that is currently pinning them in).


      All of the Intel vs ARM stuff is assuming they won’t achive the 10mm shrink. I think it is unreasonable to assume that.


      That said, drop an “A” SoC chip and LTE into the 12” MacBook and most of its compromises go away. (Also, a 10nm Intel chip fixes most of them as well. The problem with it is the severe constraints of the chassis thermals...which limits it to the low-wattage 14nm “m” chips.)

    • Avatar

      wright_is

      In reply to longhorn:

      I think that is the biggest problem. Until ARM64 apps start appearing in numbers, Windows on ARM (WOA/WOS) is going to be for enthusiasts or people with very limited needs.

      That the new version of Visual Studio arrived last week with ARM64 support might help ease matters, but projects still need to be modified to run on ARM64. With luck, it is just recompile and test, if it is all managed code, but if there is C or assembler involved, a lot of extra work will be involved.

      Then there is the chicken and egg problem, why bother re-writing / re-compiling for ARM64, when "nobody" uses it (< 10%) and why buy an ARM64 device, when there is no native software, a Pentium Gold might not be as efficient as a Snapdragon, but if it is running the code native, it is still going to be head-and-shoulders better.

      Microsoft really need to get people off of Win32 and onto modern applications, but even they haven't managed it so far with full Office. There is just too much legacy code to make such a move quickly - AFAIK, a lot of Excel functions, for example, were written in assembler for performance. You can't use assembler in UWP, so great swathes of code will need to be written from scratch and the new, managed p-code will have a performance disadvantage.

      • Avatar

        skane2600

        In reply to wright_is:

        I mostly agree but I think the WOA use is more like <1%. A lot of tech people say that MS needs to get people off Win32 but I haven't seen any analysis that shows why this is so important. IMO the "promise" that everything is going to be done on mobile is fading fast with smartphone sales saturation and tablet sales declining. The structure of the human body, not technology, is the limiting factor in making mobile practical for non-consumption activities.

        • Avatar

          wright_is

          In reply to skane2600:

          Sorry, maybe I put that poorly. I meant, as long as ARM is below 10%, nobody will buy it and nobody will develop for it. You are correct, it is easy below 1% at the moment.

          As to why it is important, because Intel keep failing to respond to mobile computing. They keep postponing the move to 10nm, whereas Huawei, Samsung and Apple are already at 7nm... They have been testing on their laurels for too long. They need another kick up the rump, like AMD gave them back in 2002.

  4. Avatar

    skane2600

    "Qualcomm’s 2019 offerings will be significantly and noticeably better than Intel’s, particularly in tablets."


    Perhaps you could claim that Qualcomm's offerings are better than Intel's specifically for tablets. But unless you are cherry-picking figures of merit that favor Qualcomm, it's not credible to claim Qualcomm is better in the general case.

Leave a Reply