Comments Locked

41 Comments

Back to Article

  • Targon - Wednesday, October 17, 2018 - link

    So, Intel felt comfortable with virtually no improvements for the past four years until AMD released Ryzen and TSMC was on the verge of getting 7nm fully up to speed. That is when stock holders and the executives freaked out, demanded an explanation for the lack of any progress, and heads started to roll. This really has been entertaining to watch over the past twelve months since Intel executives started to freak out with no plans having been in place.
  • HStewart - Wednesday, October 17, 2018 - link

    I think there is a confusion between Node development and actually functionality development.
  • shabby - Wednesday, October 17, 2018 - link

    Both have stalled.
  • Targon - Wednesday, October 17, 2018 - link

    Design improvements are where IPC improvements come from, adding useful functionality as well. Since IPC has not been improving at Intel, that implies the design stuff hasn't been happening. Fab process improvements would help with higher clock speeds, but those have not really improved in the past four years or so(their solution was to replace TIM with solder to see the true potential of the chips, which were artificially limited by the TIM in the first place. A delid shows that to be the case).

    As I said, no real improvements for the past four years at Intel, not even a core count increase. Ryzen came out, and suddenly you have Intel increasing core counts, but IPC is stil stuck.
  • KAlmquist - Wednesday, October 17, 2018 - link

    The last time Intel was facing serious competition from AMD, Intel came up with the tick-tock approach. The point of the tick tock approach was to spend a bit more on development costs in order to reduce the damage when things went wrong. It worked as intended when Intel ran into trouble with its 14nm process. Broadwell (the shrink of Haswell from 22nm to 14nm) barely happened, but Skylake made it out more or less on time. But the 14nm delay broke tick-tock going forward because even if Intel could go from 14nm to 10nm in the two year time frame assumed by tick-tock, that would still make 10nm a year late. The low risk choice for Intel would be to develop a 14nm version of Ice Lake. This would have cost money but would have ensured that Ice Lake would not be delayed even if 10nm ran into problems. I can only suppose that after seeing AMD come out with Bulldozer and Global Foundries struggles to develop new nodes, Intel figured the risk-adverse approach that led to tick-tock was no longer necessary. Then AMD came out with Zen, built on a process that Global Foundries licensed from Samsung, and Intel is scrambling.
  • Santoval - Saturday, October 20, 2018 - link

    "Since IPC has not been improving at Intel, that implies the design stuff hasn't been happening."
    We already know that, nothing needs to be "implied" : Intel has retained the exact same design since Skylake was released. The exceptions are their HEDT CPUs, where they quadrupled the L2 cache, reduced the L3 cache and added an AVX512 block, and Cannon Lake, which is the Skylake at 10nm plus an AVX512 block.

    However Cannon Lake (and 10nm) looks like it is going to be skipped, since Intel intends to release Ice Lake in late 2019, from top-mainstream to bottom. Ice Lake is their first new design since Skylake and will also be fabbed at 10nm+, which is an optimized 10nm node with many fixes on top from what they learnt from trying to fab and increase the yields of 10nm CPUs. So, in effect, Cannon Lake will remain a beta CPU and 10nm a beta node.

    Ice Lake should be fast though. It will be fabbed at their 2nd gen 10nm+ node, it will be a new architecture, it will have one or two AVX512 blocks, probably a new cache hierarchy, perhaps a new inter-core link, and possibly a redesigned front-end and back-end. Will Intel manage to release it in high volume in late 2019 though, or will it slip into 2020?
  • 0ldman79 - Tuesday, October 23, 2018 - link

    Cannon Lake has AVX512.

    Ian Cuttress did some testing, not sure if he wrote an article but he definitely talked about it on Twitter.
  • wumpus - Wednesday, October 17, 2018 - link

    It looks like all of Intel's plans were "and 10nm will be ready". Unfortunately, nobody cared about keeping the really good process guys needed to make sure 10nm was ready.

    It looks like they weren't surprised it was late, but had no idea they would be here. Now they are just rearranging the deckchairs and hoping that a miracle in the 10nm fabs will occur.
  • Eletriarnation - Wednesday, October 17, 2018 - link

    It's fun to dunk on the king, but I think it's more likely that Intel encountered actual technical difficulties with 10nm than that they just decided to fire everyone who knew anything about how to do it.
  • peevee - Wednesday, October 17, 2018 - link

    They might not have fired them - instead they did not retain them against offers from Samsung, TSMC etc (on the process side)/Apple, Qualcomm etc (on the arch side), and instead have hired a bunch of "politically correct" hires. Even if everybody would be retained, simply hiring bad people hurts team performance all by itself.

    I would not be surprised if they played with offshoring to countries which never produced anything competitive, which have hurt every company which tried it.
  • Kvaern1 - Thursday, October 18, 2018 - link

    Intel milked the cow to the bitter end, just like any other large corporation in a similar situation would do, but that being said Intel has also been seriously upgunning for a while now so I'd expect a pretty strong comeback.
  • FullmetalTitan - Thursday, October 18, 2018 - link

    100% accurate take. Intel has been bleeding talent for a couple years to all of the above. Management has been a major issue per some of those that jumped ship, and is holding them back from executing on the vision their engineers have.
  • JKflipflop98 - Wednesday, October 17, 2018 - link

    You couldn't be further from the truth. You obviously have no idea what you're talking about.
  • HStewart - Wednesday, October 17, 2018 - link

    Just keep in mind - AMD fans love to be one side and don't believe in improvement outside of Node side. Also don't want to believe that Intel 10nm is actually same as others 7nm

    There is more that makes up a CPU than node size - I remember the days of Frequency wars - and it that case it was actually Intel with raising CPU speed.
  • sa666666 - Wednesday, October 17, 2018 - link

    AMD was not mentioned once in this thread until you brought it up. You just can't resist bad-mouthing other companies when your precious Intel is being cast in a bad light.

    Look, Intel screwed up over the past few years. They got complacent and lazy. Not the first company to do it, and certainly not the last. Can you just accept that your company-worship looks extremely pathetic, and that Intel just made a mistake?? They can likely right the ship and come out of it, but your blind loyalty really makes you look like a clown.
  • Targon - Wednesday, October 17, 2018 - link

    IPC, you missed that if IPC has not improved, then where is the "new generation" that shows the design, not fab process has been worked on? No higher clock speeds from Intel in a long time(outside of increasing base/turbo speeds to get closer to the true potential of the chips), no IPC improvements, so no better performance per clock. That is where people can fairly say that Intel hasn't done much to improve their chips in four+ years.
  • peevee - Wednesday, October 17, 2018 - link

    Clock increases are basically stuck since 4GHz Pentium 4s. There is no way around physics, power consumption=heat grows as square or faster with clock while performance less than linearly (near enough to process limits that is).

    Architecturally, the limitations of both increasing clock and Von Neumann-derived architecture were hit >10 years ago, and only density improvements with SMP on chip have kept them going - which are over at so-called "14nm" (14+ and ++ are less dense, and even 32, 22 and 14nm are not really 32, 22 and 14 in any sense, transistors are not even smaller than on 45nm node, just FinFET and more precise manufacturing enabled lower voltages for the same clocks and consequently thinner wires/denser packing).

    A lot, a lot of BS was said by Intel marketing - but the truth (above) is simple.
  • FullmetalTitan - Thursday, October 18, 2018 - link

    Clock speeds are gated by simple physics. RC delay is a bitch
  • RSAUser - Thursday, October 18, 2018 - link

    Intel 10nm is definitely not the same size as other 7nm and there is an article about it somewhere on anandtech.
    Their 14nm was better than the competition, their 10nm probably won't be.
  • ToTTenTranz - Wednesday, October 17, 2018 - link

    I wonder how much of this surprise came from Intel counting on AMD to just stick with the lesser-competitive GlobalFoundries for CPUs until the end of times.

    Truth be told, if AMD was stuck with GF (and their constant inability to release top-end processes on time) then Intel would probably get another year or two with their 14++ vs. 12LP process advantage, and that could be enough to finally solve their 10nm problems.
  • Targon - Wednesday, October 17, 2018 - link

    Ryzen came out at Global, and surprised Intel with how competitive it was. Clock speeds are the only area that AMD is currently behind on. Since this article talks about Intel production, it is easy to say that Intel is well behind where it was supposed to be at this point, and heads are rolling as a result.
  • HStewart - Wednesday, October 17, 2018 - link

    This sounds to me very smart what they are doing.

    Keep new technology separate for manufacturing is smart - it like demand of existing products has hurt 10 nm development and they need to keep up with demand at same time not have existing products demand hurt new development.

    The title is a little misleading - it not really splitting manufacturing - maybe for last two cases but it basically switching R&D from production. But it also possible that "Manufacturing and Operations" has some R&D development - I would think Node development is different that actual process functionality development. For example, new functionality like Security fixes in hardware should not required 10nm.
  • JKflipflop98 - Wednesday, October 17, 2018 - link

    So basically we're going back to doing what we were doing six years ago before the Ramp/PTD merge. Great.
  • CaedenV - Wednesday, October 17, 2018 - link

    Granted, what they were doing 6 years ago worked, so this is not a terrible idea.
  • SharpEars - Wednesday, October 17, 2018 - link

    Ian, do you spellcheck your posts before putting them in front of millions of people?

    "producion"??? Seriously?
  • Ian Cutress - Wednesday, October 17, 2018 - link

    Written on a smartphone while on the move. If you hate my speed spelling, you should read my live blogs.
  • hanselltc - Thursday, October 18, 2018 - link

    Just a curious question: do you not bring a keyboard around? Before I had my laptop for college I bought my K63 around for doing essays and takign notes with an iPad.
  • SharpEars - Wednesday, October 17, 2018 - link

    Also, it's "hit primetime" not "hit the primetime."
  • Ian Cutress - Wednesday, October 17, 2018 - link

    A beg to differ on this one. Perhaps it's a regional idiomatic thing.
  • Farfolomew - Monday, October 22, 2018 - link

    Back off buddy! 'Your' being annoying ...
  • bill44 - Wednesday, October 17, 2018 - link

    10nm: "The process technology was originally set to have been in production in 2016"

    "Normally we expect to see a new major manufacturing process every 18-36 months"

    If 10nm was expected in 2016 & new major manufacturing process every 36 months, 7nm probably was planned for 2019.

    Is it possible 7nm will be ready before 10nm? Or, 10nm will be short lived?
  • A5 - Wednesday, October 17, 2018 - link

    IIRC, Intel always had 7nm as their first EUV node. I would guess that is just as delayed as 10nm has been.
  • HStewart - Wednesday, October 17, 2018 - link

    Intel could just rename 10nm to 7nm - which is known that Intel 10nm >= others 7nm anyway.
  • sa666666 - Wednesday, October 17, 2018 - link

    'Known' to you, at least. You do realize just because you keep repeating something doesn't make it true, right? What you're saying is entirely supposition, and you have no proof whatsoever that this is true. But because it fits your worldview, you keep on repeating it.

    I think that's why many people here find you annoying. Not because of you eternal love for anything Intel (although that is annoying in an OCD way), but because you keep repeating stuff not known to be facts as if they are absolutely true. You're talking out of your ass much of the time, and don't really have a solid understanding of what you're spewing.
  • hanselltc - Thursday, October 18, 2018 - link

    Pretty sure it doesn't take 10 minutes of googling to find out what happaned to nodes other than intel's when finfet became a thing?
  • Lord of the Bored - Wednesday, October 17, 2018 - link

    [citation needed]
  • msroadkill612 - Wednesday, October 17, 2018 - link

    If amd were making 7nm bulldozers now, they would be uncompetitive against 14nm.

    The fact remains, that monolithic and ring/mesh bus architecture chips have been shown to not be the competitive way forward from here.

    The writing was on the wall long ago, but intel thought it could dictate to the market what suited intel, and now its too late.
  • GreenReaper - Thursday, October 18, 2018 - link

    So, uh, what's your alternative? Last I heard Threadripper had pretty high interconnect power usage too. I like the cut of AMD's jib, but more cores aren't free for either party; for most use-cases four or six will be plenty. It's certainly nice to have higher-end options to bring the price down, though.
  • Buk Lau - Wednesday, October 17, 2018 - link

    F
  • PeachNCream - Thursday, October 18, 2018 - link

    G
  • Ananke - Thursday, October 18, 2018 - link

    Intel is not competing with AMD only, it is competing with NVidia as well. But these two examples are fabless...
    Where Intel was falling behind as primarily being dab manufacturer, is TSMC and especially Samsung. Today, Samsung announced they are moving production to ExtremeUV process, which allows them to work on 7nm and below nodes. That's Intel's problem...Samsung already ate Intel's share in memory business /Intel became what it is as a memory maker/, eventually Samsung will start getting the processors fab business as well.

Log in

Don't have an account? Sign up now