back to article Chin up, Intel heads - you can still lop legs off ARM's data-centre dash

A new CEO – Brian Krzanich – and president – Renee James – have taken the helm at chip giant Intel, although they have yet to articulate a grand vision for Chipzilla for the next decade. But it looks like we may be getting some sense of Intel's long-term plans in the data centre at an event that the company is hosting on 22 …

COMMENTS

This topic is closed for new posts.
  1. IGnatius T Foobar
    FAIL

    Microsoft FAIL

    (Subject line in homage to the One True Commenter, of course.)

    As the article pointed out, Windows Server doesn't run on ARM, nor do Microsoft's dodgy back office applications such as Exchange and their broken port of Sybase. The solution to this is of course running Linux software which is portable to every architecture. ARM servers are coming whether Intel likes it or not.

    1. Tom 7

      Re: Microsoft FAIL

      If one of my minions came to me and said he couldn't get his code to run on ARM I would be up in front of an industrial tribunal explaining why I cruelly took the piss out the said employee and that would provide the seed fund for their training in some job they could do like emptying the bins or sales or management as they have the lying skills to get that far in IT.

      When MS cant get their code to run on ARM then I advise everyone to migrate away from them ASAP. It will save a fortune in the short term and a complete loss of all your data in the long term.

  2. Roo

    It is hard to take TPM's speculations seriously

    It is hard to take TPM's speculations seriously, here's why. :)

    Historically Intel used to license the masks so that other vendors could act as a second source for Intel parts. That is how AMD got into the game - although Intel have tried to revoke their license on more than one occasion. Licensing the masks would be useless these days because Intel's processes are very different from those available to the rest of the world.

    So that leaves us with licensing the ISA and designing cores which is very expensive (providing you can get the requisite skills in the door or develop them in house). This has not deterred folks in the past (eg: VIA, AMD, IBM, Transmeta et al), but today only AMD has any significant market presence today. The high rate of fail suggests that it's very hard to make enough money from a clone of an Intel part.

    I think Intel will have to license the ISA and ready-to-fab cores to gain any traction at the expense of ARM, if they fail to do that the high barrier to entry of designing cores will hamstring their 'ecosystem'. Even if they do license the ISA and core designs are there are enough folks who know how to design Intel based SoCs out there ? Also I suspect that the Intel solution will be more costly to produce (die area, yield) and validate because the 'x86' ISA is far more complex than the ARM ISA(s).

  3. John Smith 19 Gold badge
    Unhappy

    "Xeon becomes the new Itanium"

    I wonder if the Xeon team are looking forward to that idea?

    I doubt MS will pay anything but lip service to the idea of porting a public version of Windows to ARM.

    Their ongoing and deeply dysfunctional relationship will continual and they will remain the Fred & Rose West of IT for another decade.

  4. Jim O'Reilly

    Is Wintel the next "mainframe"?

    To be a winner, you have to provide enough value to keep customers loyal, and occasionally do a brilliant end run on the competition.

    Intel needs a high-core-count low power solution with virtualization that runs like an ARM. They've got such a product in their Lab, with 80 cores and a 62W profile. That could be an ARM-buster.

    But x86 and Windows have a problem. On the one hand an expensive and perhaps tired OS (getting terrible press on the desktop) and on the other a complex architecture that either costs a lot or uses a ton of power.

    If the ARM64 supports KVM well, that may become the vitualized high-core-count server of the future, with a very large premium for VMware, Windows and x86.

    This will be a much bigger issue if HMC memory melds with those ARM devices. A single module with KVM and a terabyte of DRAM, mounted in a micro-server chassis with 45 modules? That's potentially 256 VMs per module and 11,000 VMs per server box.!

  5. mark l 2 Silver badge

    I expect that MS have a alpha ARM port of Windows server 2013 in the labs just in case the ARM servers running linux begin to really take off. After all Windows server and Windows 8 use the same underlying code and Windows 8 runs on ARM with the RT version so essentially its just the server services and extra programs and admin tools that need to be compiled for ARM to get a working version.

    1. John Smith 19 Gold badge
      Unhappy

      @mark 12

      "I expect that MS have a alpha ARM port of Windows server 2013 in the labs just in case the ARM servers running linux begin to really take off."

      I can easily believe that MS have a lot of stuff in their labs that will never see the light of day.

      The days when whole OSes are written in assembler and porting them in any realistic timescale to any other architecture (except one that is 99% opcode compatible) is long past.

      1. lightknight

        Re: @mark 12

        Yes, and no. Remember, when it comes to programming, there are always people trying to speed things up with a few architecture dependent tricks. Because that's what people want...and businesses want....right up until they need to port the code to a new architecture....at which point they prize that.

  6. bazza Silver badge

    Intel's has one ultimate problem: the X86 instruction set takes too many transistors to implement.

    Everything about the modern X86/X64 architectures from both AMD and Intel shows just how clunky the instruction set is. The pipelines, instruction decoders, branch predictors and caches that are needed to make x86 perform well take an enormous number of transistors. Both Intel and AMD have done really well to make it work as well as it does, but the efficiency is low.

    Whereas the ARM instruction set is far simpler to implement to perform well. Memory, modest cache, simple pipeline, CPU core, job done. That means less transistors, which means less power, etc. etc. The only reason Intel are still 'ahead' is because they're so good at making transistors, disguising their architecture's profligate use of them.

    So as soon as Intel lose that silicon processing edge, they're doomed. ARMs can nip in and perform just a little bit better, and that's the server market gone ARMwards. There's not a lot they can do about that. The world didn't like Itanium. I doubt there's room for yet another CPU architecture. They could build their own ARM based CPUs, and they'd likely be the very best that you could buy.

    For those of you muttering about MS being left behind - don't forget that at some tech shows some years back they showed desktop Windows working on an ARM board running Office and printing. If that's where they could get to with an experimental Alpha build then I don't suppose that porting Windows server is going to be that hard. Probably the biggest portion of the job is verification & testing.

    1. Anonymous Coward
      Anonymous Coward

      "Probably the biggest portion of the job is verification & testing."

      I presume you weren't around last time MS tried to make Windows multi-platform?

      Verification and testing of MS code may be tedious for MS, but it wasn't a problem for the 3rd party software writers last time round. They just ignored anything that wasn't x86 (meaning they ignored Alpha, and MIPS, and PowerPC). Look up Advanced RISC Computing, back in the days of Windows NT.

      Lots of things have changed since then, but I don't see what would now motivate the third parties to port to a non-x86 variant of Windows when they didn't do so before. I think most sensible vendors with finite resources might prefer to use them to go with the market and look at getting off Windows altogether, rather than simply extending the inevitable decline of Windows.

      Apple or Android or Linux next, that's the real question for these folk.

  7. Anonymous Coward
    Anonymous Coward

    You're having a laugh. Intel have interesting times.

    ARM CPUs were built from the start to be efficient and cheap, so no way can Intel compete unless they scrap the x86 architecture.

    Even the cheapest AMD APUs thrashs the Atom for performance, and mobo support, and work out cheaper too, a 64-bit ARM APU should be a lot cheaper, and finally make the Atom obsolete.

    If AMD do build 64-bit ARM SoC APUs and get server makers on-board, this could cannibalise Intels low end CPUs, for server and client machines. Nvidia and other ARM makers may also join the fray and put yet more pressure on Intel.

    Windows is not need for server and casual use, and this will cost Intel and Microsoft increasing market share.

  8. dmcq
    Thumb Down

    Missed the point

    There's no point Intel licencing its CPU designs, who'd want them? Intel itself would be too much competition. And it would be attacking its cash cow. Their concentration on the CPU makes them think ARM is the competition. What it should have done long ago is open its foundries and deter the other foundries from competing at the high end. However its foundry plans looks like it is just trying to use unused capacity. Doing it properly would take a couple of years even if they really wanted to and anything can happen by then - I think the technology gap is closing as the foundries all come against some hard limits.

This topic is closed for new posts.

Other stories you might like