back to article A bridge over troubled water: Intel teases Ponte Vecchio, the GPU brains in US govt's 1-exaFLOPS Aurora supercomputer

Intel today confirmed the identity of the GPU-based math-accelerator chip it will offer to supercomputer builders. The announcement of the Ponte Vecchio part, named after that medieval bridge in Italy, coincided with the start of the Supercomputing 2019 conference in Colorado, USA, on Sunday. Ponte Vecchio will be fabricated …

  1. Anonymous Coward
    Anonymous Coward

    I assume there'll be a forthcoming build of Windows 10 to throttle these insane specs down to something we can all appreciate.

    1. Phil Kingston
      Thumb Up

      I salute you - that's the biggest reach-out from an unrelated article to crow-bar in an anti-Windows comment I've seen for quite some time.

      1. Anonymous Coward
        Anonymous Coward

        It’s literally like Hitler and Brexit.

        1. MiguelC Silver badge

          That literally doesn't mean what you think it does.

        2. Mike 16

          Godwin and Boris

          --- It’s literally like Hitler and Brexit. ---

          Wasn't the Leader more like in favor of a Bunion (Britain in Union)? I mean the one who steamrolled Chamberlain, not the one who has been renewing the fire insurance on the houses of Parliament.

        3. StuntMisanthrope

          Gotcha.

          The last time a speaker interfered in politics ended well. #hindenburg

  2. sbt
    Paris Hilton

    We'll burn that bridge when we come to it

    Serious question; what's different about Intel's 7nm process that leads to confidence they can deliver it after what happened with the 10nm?

    1. Pascal Monett Silver badge

      My thoughts exactly. They haven't ironed out 10nm, how can they possibly get 7nm right ?

      Or is it a case of 10nm is bolloxed so let's just leave the mess and move to 7nm ?

      1. Anonymous Coward
        Anonymous Coward

        "Or is it a case of 10nm is bolloxed so let's just leave the mess and move to 7nm?"

        Pretty much this - Intel 10nm has a lot of unrelated issues, some of Intels making, some of the equipment that they were using was not able to deliver the speed or accuracy required for a reliable process in addition to process issues (too many quad patterning steps that resulted in very tight tolerances that equipment couldn't deliver reliably) and chemistry issues moving to cobalt interconnects. They weren't necessarily the wrong steps for 10nm, but doing them all at once meant that the risk of very complex interactions that were not understood if things went wrong were high.

        The problem for Intel was that by the time the issues with their 10nm were understood and there was equipment to fix the issues (i.e. the second gen equipment that TSMC is using for their 7nm process which is comparable in that it also uses some SAQP steps) it would mean ANOTHER 2 year delay to get it built/installed.

        I believe Intel took the option of both new 10nm equipment (they will still need a node between 14nm and 7nm for future products) and ordering 7nm equipment for new fabs to stay competitive with TSMC/Samsung.

    2. analyzer

      Re: We'll burn that bridge when we come to it

      Remembering that Intel measure size differently, their 10nm is equivalent to the rest of the industry stating 7nm, 7nm will be 5nm and 5nm will be 3nm. Not confusing at all no siree,

      Intel tried to stay with DUV for 10nm, every other major silicon scratcher went for at least 2 layers of EUV. Intel had to restart their 10nm and, lo, just like the rest of the industry it will use EUV in small part. Their 7nm and 5nm processes are on schedule because they have planned EUV in just like everyone else. They are still behind TSMC who are already researching 1nm, will be introducing 6nm next year, 5nm in 2021 and 3nm in 2022ish with 1nm by 2024.

      Intel is behind in process and production capability for at least another 6 to 12 months, what happens to the market in this time could be interesting.

      1. IJD

        Re: We'll burn that bridge when we come to it

        The TSMC process dates are the wrong way round -- 5nm (lots of EUV) is first, then 6nm (which is a shrunk 7nm with 1 more EUV layer), then 3nm. Exact dates depend how you define them, I believe some customers already have working 5nm silicon.

        Either way there's no doubt Intel is well behind, at least 12 months even if things go well for them on 10nm (7nm TSMC) and 7nm (5nm TSMC) -- they really dropped the ball on 10nm and managed to convert a 1-2 year process lead into a 1-2 year lag, which takes some doing...

      2. Anonymous Coward
        Anonymous Coward

        Re: We'll burn that bridge when we come to it

        They are still behind TSMC who are already researching 1nm ...

        For silicon, 1nm is about six atomic layers[*]. My mind boggles.

        [*] The silicon-silicon single bond is about 0.23nm; however, in the diamond-like crystal structure of silicon, the layers are spaced on average by 2/3-rd of the bond length.

      3. Anonymous Coward
        Anonymous Coward

        Re: We'll burn that bridge when we come to it

        TSMC stuck with DUV for low/high performance 7nm with a few SAQP steps.

        TSMC is introducing EUV with 7nm+ and rumoured yields for low performance 7nm+ are not great at present and even worse for high performance 7nm+.

        Samsung have taken a similar approach - while they had planned for EUV, delays to EUV meant their current 7nm node (7nm ArF) is using DUV with an updated 7nm EUV process available this year, although I have not seen any yield/performance data.

    3. Persona

      Re: We'll burn that bridge when we come to it

      Different teams worked on 7 and 10nm ...... so I'm betting the 10nm team is really hoping the 7nm guys screw up too.

      1. Tom 7

        Re: We'll burn that bridge when we come to it

        ISTR people doing this actually enjoy the challenge and the 10nm team will be hoping the 7nm team get it almost right so they can leapfrog to whatever comes next.

        I am prepared to believe modern management may insist on people not enjoying their job but there is some progress here so I guess the people in the know keep them out of arms reach.

      2. Anonymous Coward
        Anonymous Coward

        Re: We'll burn that bridge when we come to it

        "Different teams worked on 7 and 10nm ...... so I'm betting the 10nm team is really hoping the 7nm guys screw up too."

        If Intel screw up 7nm, it will hurt Intel badly and the fab/CPU market will likely change fundamentally.

        It would likely mean significant consolidation as Intel will drive a lot of the smaller fabs relying on older nodes out of business as they search for revenue to fund "next-gen"

        At the high end, TSMC's capacity constraints would likely limit production significantly and completely alter where many products are produced as constrained capacity would likely lead to high margin kit like x86 CPU's (AMD), GPU's (nVida/AMD) dominating and low margin/high volume chips (non-premium ARM chips) moving to other fabs.

        1. Terje

          Re: We'll burn that bridge when we come to it

          Nvidia is supposedly moving manufacture of Ampere to Samsung from TSMC, from what I read capacity constraints of TSMC would likely be a significant factor in that.

          1. Anonymous Coward
            Anonymous Coward

            Re: We'll burn that bridge when we come to it

            I'm not sure if it was capacity constraints or concerns around the Global Foundries lawsuit that has since been resolved.

            I suspect nVida used to be the top dog amongst TSMC's customers and first Apple and now AMD have moved ahead of them. Samsung is less about capacity (for now) and more about business options in the face of a changing market.

            That may change if AMD's CPU market share increases significantly but I think we are still ~6+ months away from being able to judge that as I suspect we will get announcements once it has happened rather than pre-announced as the market in 2020 is uncertain.

          2. Anonymous Coward
            Anonymous Coward

            Re: We'll burn that bridge when we come to it

            The "moving to Samsung" claim has been floating around for at least four years.

            All of the factors mentioned are parts of the reason, but that doesn't mean the switch is going to happen, or that it will be complete.

            Arguing against the change is that Nvidia is extremely single-threaded: they have only one design pipeline. That makes them very efficient, but it means that any failure is a crisis. Two failures in a row could be a company-breaker. That makes them very reluctant to redesign problematic parts of their architecture, instead preferring a series of half fixes. It also means that they aren't likely to suddenly switch to a new set of design rules for a new fab. Perhaps four years of rumors means that they have been working on it as a side project. But even if that's the case, the same conservative approach means that they are unlikely to make a sudden switch.

            1. Terje

              Re: We'll burn that bridge when we come to it

              What makes it more likely now compared to the last couple of years to my mind, is that it aligns with a process shift as well, so they still have to do the redesign of existing parts for a the new smaller process. This should make the extra work/risk of shifting to another manufacturer as small as possible.

    4. sbt
      Pint

      Getting straight the ponte

      Thanks for all the answers here so far; some fascinating insights. Thumbs all around.

      It's threads like these that have me questioning the "commentard" label. More like "commentoffins"!

      1. Anonymous Coward
        Anonymous Coward

        Re: Getting straight the ponte

        It doesn't mean quite what you think, it has nothing to do with that awful American insult "retard".

        The members of the Paris Commune that briefly controlled the city after the Franco-Prussian war in 1871 were called "communards". Hence "commentards" - members of the group of people who comment.

        - Fier commentard depuis 2010, mais sous queleques nommes de plume clavier.

        1. Anonymous Coward
          Anonymous Coward

          Re: Getting straight the ponte

          I think you are being overly generous to the ElReg community.

          Pick the appropriate article (Trump, Brexit, Windows patching issues, Android Vs iPhone, tabs Vs spaces) and all hell breaks loose.

          "Commentard" is accurate for myself at least - hence the cowardice in case I'm drawn into correcting someone who is wrong on the internet and I'm worried my friends or workmates may recognise me....

    5. ToddRundgrensUtopia

      Re: We'll burn that bridge when we come to it

      three

  3. Anonymous Coward
    Anonymous Coward

    It will be a supercomputer until more Intel security fuck-ups crawl out of the wood work then after mitigations are applied it will then have the speed of a VIC 20.

  4. StargateSg7

    BEAT THIS INTEL, AMD AND ARM !!!

    128-bits wide 60 GHz GaAs Super-Workstation/Super-Server CISC Chip At Final Tape-Out Stage!

    https://www.reddit.com/r/Supercomputers/comments/dx4c74/128bits_wide_gaas_superworkstationsuperserver/

    We Win !!!

    .

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon