back to article When the chips are down, Intel's biggest gamble isn't what to do – it's whom to do it with

Political America likes to judge its presidents by their first 100 days. Corporate America thinks more in 90-day cycles, so as today is Pat Gelsinger's 90th day at the helm of Intel, it's an apt time to look at how he and the company are doing. The market remains cool on Gelsinger, perhaps because he said that the $80bn of …

  1. IGotOut Silver badge

    Subsidies?

    Remind me how much PROFIT, they make?

    And no doubt, the tax owed selling those parts, won't be paid in those places, y'know, IP rights and all that.

    1. Flocke Kroes Silver badge

      Re: Subsidies?

      The alternative to subsidies is to set the minimum wage to under $100/month, an enormous supply of free water an no limits on pollution.

      1. Yet Another Anonymous coward Silver badge

        Re: Subsidies?

        And guaranteed customers.

        Intel build a cheap "CPU factory" that packages chips made by TSMC - call it encapsulation of core modules to confuse politicians.

        Require that US federal funded agencies buy these "Made in America" chips for reasons of national security.

        Charge twice as much for them

        Profit

        1. bombastic bob Silver badge
          Devil

          Re: Subsidies?

          Require that US federal funded agencies buy these "Made in America" chips

          That (inevitably) may be necessary at some point... And hopefully it makes pure economic sense, especially if the process is heavily automated.

          1. Yet Another Anonymous coward Silver badge

            Re: Subsidies?

            What it will actually mean is that if your university lab gets a NSF grant you have to throwaway(*) everyone's Lenovo laptop and buy them all Macbooks with "Assembled In USA" approved chips.

            * actually you can't throw them away because the paperwork of disposing of any computer that was bought on one grant and use don another is so complex that even string theory doesn't have a potential solution. At some point we are going to have to get a grant to pay for the storage of old IT equipment that can't be disposed of.

  2. Anonymous Coward
    Anonymous Coward

    Sooner

    Gelsinger should have got the job 5yrs ago.

  3. Anonymous Coward
    Anonymous Coward

    sour grapes marketing

    Intel's apologists love to repeat the mantra that size isn't everything ("Intel's 10nm is better than TSMC's 7nm!"). This is a variant of the same thing we saw in x86 processors many years ago when Cyrix, who had some interesting architectural features but were unable to match Intel clock-for-clock, came up with the ill-fated "Performance Rating". The PR was some multiple of their actual clock rate intended by marketing types to be compared with others' clock frequencies. Cyrix weren't the only ones to do this; AMD dabbled and so did others. Of them all, only AMD are still in business and they've been transparent about their speeds and feeds for many years now.

    It's absolutely true that power, performance, and price are what actually matter. And that's why this claim simply doesn't hold water: Intel still can't produce enough "Intel 10nm" parts to meet demand, and neither those nor the "Intel 14nm" parts that make up the bulk of their offerings are even remotely competitive. Yields on the 10nm process are terrible, requiring their monolithic wafers to be heavily binned and resulting in core counts roughly half the competition's. Despite having fewer cores and tiny caches, those processors dissipate more power than both AMD's and Ampere's. Rounding out the trifecta, the top of the line Xeon lists for 20-50% more than AMD's superior 7742 and 7763. Cascade Lake wasn't competitive with Naples, even while AMD shipped Rome. Ice Lake isn't competitive with Rome, even as AMD ship Milan. Intel are losing on every imaginable metric, even the niche metrics they've relied upon in trying to gloss over their failure. Whether they're losing because "Intel 10nm" really isn't better than "TSMC 7nm" is immaterial, but one can't support the "process size doesn't matter" argument with price, performance, or power figures. All three suggest that process matters and TSMC's is probably quite a bit better. That's further bolstered by the fact that Ampere, using the same TSMC process as AMD to build processors of an entirely different ISA, are coming out with even higher core counts than AMD (to be expected, honestly, given the ISA differences) and similar performance figures at the same power levels as AMD. In other words, both the leading large-processor manufacturers using TSMC's process are kicking the snot out of Intel on every metric.

    Sour grapes marketing doesn't work. People know that 7nm is better than 10nm, and if there were any doubt about it, they can look at the things that do matter and see that there is, if not a causal relationship they can understand physically, at least an overwhelming correlation. Please stop propagating this nonsense. If Intel want to tell us that nominal process node doesn't matter, they're welcome to prove it by delivering a competitive product on their "in some ways better" process. They've been trying to do just that for more than 5 years now and not only haven't caught up but continue to lose ground . Maybe Intel 7nm -- if they ever get it working -- will deliver better processors than TSMC 5nm, but history suggests it probably won't, and in the meantime I cringe every time I read this nonsense. The grapes are delicious; don't deprive yourself.

    1. EnviableOne Silver badge

      Re: sour grapes marketing

      The problem for Intel is their x nm may be as good or better than TSMCs next step, but TSMC are at gen 2steps ahead so still outperform in the key metrics (Speed, power consumption, and price)

  4. Duncan Macdonald
    Mushroom

    Intel has major design problems

    All of Intel's CPUs for many years have been monolithic (everything on one piece of silicon) with structures such as their ring bus providing a significant part of their improved performance compared to the pre chiplet AMD designs.

    To match the core counts that AMD can manage in a single package, Intel will need to move to a chiplet design to get acceptable yields. However a chiplet design will mean discarding much of the "secret sauce" such as the ring bus that provided the improvement relative to AMD's monolithic designs.

    Intel need to do a MAJOR architecture redesign to use chiplets while at the same time trying to get their basic silicon speed to match what TSMC will be able to get from their 3nm node.

    Intel will also have another major problem - patents - I would be willing to bet that TSMC has a lot of patents covering parts of their 7nm and smaller processes - Intel will either have to work around the patents (entailing extra delay) or pay TSMC to allow Intel to use them.

    Basically Intel screwed up big time by paying out money in shareholder dividends instead of investing in the required EUV based production to handle smaller nodes. Because AMD was not an immediate threat, they cut back on R&D and like all high tech firms have now found out that doing so is a long term kiss of death.

    To add to the above problems, Intel will have stock market problems if they cut their dividends to try to finance increased R&D - investors do not like seeing dividend cuts and are often far more focused on the short term rather than the long term health of a company.

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

Biting the hand that feeds IT © 1998–2022