Stories
Slash Boxes
Comments

SoylentNews is people

posted by martyb on Saturday July 25 2020, @03:07AM   Printer-friendly
from the how-the-mighty-has-fallen-behind dept.

Intel's 7nm is Broken, Company Announces Delay Until 2022, 2023 (archive)

Intel announced today in its Q2 2020 earnings release that it has now delayed the rollout of its 7nm CPUs by six months relative to its previously-planned release date, undoubtedly resulting in wide-ranging delays to the company's roadmaps. Intel's press release also says that yields for its 7nm process are now twelve months behind the company's internal targets, meaning the company isn't currently on track to produce its 7nm process in an economically viable way. The company now says its 7nm CPUs will not debut on the market until late 2022 or early 2023.

[...] On the earnings call, Intel CEO Bob Swan said the company had identified a "defect mode" in its 7nm process that caused yield degradation issues. As a result, Intel has invested in "contingency plans," which Swan later defined as including using third-party foundries. The company will also use external third-party foundries for its forthcoming 7nm Ponte Vecchio GPUs, the company's first graphics chips. Ponte Vecchio comes as a chiplet-based design, and Swan clarified that production for some of the chiplets (tiles) will be outsourced to third parties. Swan noted the GPUs will come in late 2021 or early 2022, portending a delay beyond the original schedule for a 2021 launch in the exascale Aurora supercomputer.

[...] Intel's first 10nm desktop CPUs, Alder Lake, will arrive in the second half of 2021.

See also: Intel Reports Q2 2020 Earnings: Data Center Sales Fuel Another Record Quarter
Intel 7nm Delayed By 6 Months; Company to Take "Pragmatic" Approach in Using Third-Party Fabs
Intel Roadmap Update: Alder Lake In H2'21, Ice Lake-SP Late This Year


Original Submission

 
This discussion has been archived. No new comments can be posted.
Display Options Threshold/Breakthrough Mark All as Read Mark All as Unread
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
  • (Score: 2) by RamiK on Saturday July 25 2020, @01:46PM (4 children)

    by RamiK (1813) on Saturday July 25 2020, @01:46PM (#1026152)

    It will be interesting to see what they do if AMD pulls ahead in single-threaded performance with Zen 3.

    Intel's single-threaded advantage is so tiny and limited to specific workloads and chips while coming at such a high dollars and power cost that I just couldn't care less.

    I think something like 8+32 would be better (40 cores around the size of 16 big cores), along with dropping the graphics to the minimum number of execution units necessary.

    I doubt there's demand for this sort of build server configuration that can't be adequately met by AMD or even ARM at a lower price point.

    --
    compiling...
    Starting Score:    1  point
    Karma-Bonus Modifier   +1  

    Total Score:   2  
  • (Score: 2) by takyon on Saturday July 25 2020, @02:22PM (3 children)

    by takyon (881) <reversethis-{gro ... s} {ta} {noykat}> on Saturday July 25 2020, @02:22PM (#1026161) Journal

    Intel's single-threaded advantage is so tiny and limited to specific workloads and chips while coming at such a high dollars and power cost that I just couldn't care less.

    As long as Intel's chips remain the top gaming chips, AMD loses on potential design wins and customers. Zen 3 is AMD's chance to turn that around and completely dominate for months. Intel may hang onto it by backporting architectural IPC improvements to "14nm" Rocket Lake while keeping clock speeds high. We'll know within a few months.

    I doubt there's demand for this sort of build server configuration that can't be adequately met by AMD or even ARM at a lower price point.

    I'm proposing 8+32 for consumer desktop chips. Workstation chips already have more cores (~28 big cores, or 64 from AMD). 8 big cores takes care of most single-threaded tasks and gaming, since next-gen consoles will have 8 cores, and the small cores can handle any well parallelized task, like video editing, 7zip, or whatever.

    Intel should also put out something like a 128-core or 256-core Atom chip for servers, using chiplets if necessary. Each Atom core takes up about 1/4 the die area of a regular core, but has more than 1/4 the performance.

    Intel already dabbled in high core count Atom chips... the cancelled Xeon Phi line. But those had AVX-512, which could be removed, and Atom cores are faster than they used to be (each core will match or exceed Ivy Bridge core performance in a generation or two).

    I'm just spitballing ideas here. The world doesn't really need Intel to exist for much longer, and AMD can't rest on its laurels since it is increasingly competing with ARM designs (from supercomputers to servers to laptops).

    --
    [SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
    • (Score: 2) by RamiK on Saturday July 25 2020, @03:33PM (2 children)

      by RamiK (1813) on Saturday July 25 2020, @03:33PM (#1026172)

      As long as Intel's chips remain the top gaming chips

      Considering a $200-300 CPU won't bottleneck a gaming build even paired with a $2000 GPU, I fail to see how anything but the price point determines the top gaming chips.

      I'm proposing 8+32 for consumer desktop chips...

      My definition of a consumer desktop is browsing, watching cat videos and word processing / spread sheeting so I fail to see how a 4+8 won't be more than enough. Well, that's as long as speculative execution won't have the x86 switch to real hardware threads which might very well happen... But I'd like to think we won't be using an x86 by then.

      --
      compiling...
      • (Score: 2) by takyon on Saturday July 25 2020, @11:44PM (1 child)

        by takyon (881) <reversethis-{gro ... s} {ta} {noykat}> on Saturday July 25 2020, @11:44PM (#1026352) Journal

        AMD has regularly trailed Intel in single-threaded and gaming performance for about a decade (Zen 2 wins in IPC but can't clock as high. IPC doesn't scale linearly with clock speed, and Intel does better than AMD when it can get to ~5 GHz territory. Intel's monolithic CPUs also have lower memory latency.). I think you underestimate the significance of that, even if only from a marketing standpoint. It doesn't have to be that way anymore and Zen 3 might finally put AMD on top this year. If so, it would be a gut punch for Intel and would probably force them to lower their prices, which is good for consumers.

        According to leaks [wccftech.com], Intel's Alder Lake is supposed to top out at 8 big and 8 small cores. It might come out up to a few months before AMD's Zen 4 but will likely be crushed by Zen 4 when it does come out, as AMD is already putting out 16-core "mainstream" desktop CPUs and might bump that to 24 cores on Zen 4 [pcgamesn.com]. Alder Lake could be a "Bulldozer moment" for Intel if operating systems don't handle x86 big.LITTLE scheduling well. And it's on "10nm", so if there are lingering problems with the "10nm" process, they will be in full view.

        If AMD is putting out a "mainstream" 24-core version of the 16-core Ryzen 9 3950X by 2022, or even just a faster 16-core, an 8+32 heterogeneous CPU might have a chance of fending it off. But that is not what the leaks suggest. They point to configurations like 8+8, 6+0, 2+8, and now 6+8. Even an AMD 12-core could smoke those.

        ARM is competing against x86 harder than ever before, but x86 is probably here to stay for another 5-10 years if AMD keeps crushing it. One thing that I missed: remember AMD's K12 ARM CPU [wikipedia.org] that was supposed to come out around the same time as Zen but got shelved? A recently leaked roadmap [notebookcheck.net] has "K12 FFX" on it.

        --
        [SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]
        • (Score: 2) by RamiK on Sunday July 26 2020, @06:55AM

          by RamiK (1813) on Sunday July 26 2020, @06:55AM (#1026485)

          I think you underestimate the significance of that, even if only from a marketing standpoint.

          But that's the thing: There's always some synthetic benchmark you can win. But, the thing is, there's really no single threaded loads like Intel measures them in the real world. People playing fairly casual games have a browser open in the browser, possibly with a wiki page pushing ads via javascript and a p2p app throttling a few connections so that alone is a more than a couple hardware threads and easily over a dozen soft threads running in the background.

          (Zen 2 wins in IPC but can't clock as high. IPC doesn't scale linearly with clock speed, and Intel does better than AMD when it can get to ~5 GHz territory. Intel's monolithic CPUs also have lower memory latency.).

          Look even if any of this wasn't just pure synthetic results, it's still not the bottleneck. The margin for when the GPU stops being the issue and the CPU becomes the problem is HUGE. Just 3-4 years ago I've seen mining boards working 3-4 GPUs on full throttle using i3s and over a dozen using i5s.

          Really, it all just doesn't matter.

          --
          compiling...