Comments Locked

12 Comments

Back to Article

  • PeachNCream - Thursday, December 13, 2018 - link

    I was under the impression that its well-known that Intel uses older, mature nodes to produce ICs other than CPUs. Chipsets have long been the domain of last generation fabrication and that is especially useful in desktop hardware as heat output and power efficiency are not constrained by small device form factors and limited energy storage.
  • shabby - Thursday, December 13, 2018 - link

    Were they always made at 22nm? Some sites a month ago reported the die being much larger than the current one, meaning they started using an older process.
  • mammothboy - Thursday, December 13, 2018 - link

    They do, but 14nm was supposed to be "older, mature node" compared to 10nm. With the issues there, things got jammed at 14nm. I imagine they would have pivoted back to 22nm sooner had they really appreciated how bad things were at 10nm.
  • DanNeely - Thursday, December 13, 2018 - link

    They have. The top level chipset (historically the north bridge, now the PCH with there no longer being separate north/soutbridges) has historically been about one process behind; with lesser chips using even older ones (ie the southbridge when the chipset was 2 chips). I'm not sure how old the processes for Intel NICs or thunderbolt controllers are. The RF part of wifi modules needs a different process and other than general chip making machinery wouldn't be running on old CPU tech.

    What I think happened is that the 14nm chipsets we got were designed to go with the 10nm CPUs we haven't; but that DMI was generic enough that they still worked and thus were paired with the stuck on 14nm CPUs to keep supported features coming forward until growing CPU die sizes meant they were starting to run out of capacity.
  • namechamps - Thursday, December 13, 2018 - link

    They do. However it isn't always exactly one behind. As they prep for the CPU moving to the next process node they move chipsets up to the current node as it is very mature with very high yields. It is a way of squeezing a bit of cost off the chipsets.

    So if things went according to plan (well at least the 5th revision of the "plan") Intel moves chipsets to 14nm in 2018 as it begins moving CPUs to 10nm. Except that didn't happen and now they are facing a capacity shortfall in 14nm. So they made the B365 on 22nm to free up some 14nm capacity. It is just more evidence on how atypical of a clusterfuck the 10nm transition has become.
  • HStewart - Thursday, December 13, 2018 - link

    With Technology changing and lots of Chipset functionality actually moving into CPU chip, I would expect die would be reduce. Since IO does not appear to performance - it would be logical for Intel to do this

    Also I believe Intel mention that is part of their plan - to make more time critical components like CPU, GPU and Ram using the faster process and let IO be on less expensive process. EMiB is one example of using different process on same product.

    Also 10nm is not dead, it is live and well and planned for 2019 but 14nm is still in production. Likely once 10nm and future 7nm processes are full swing 22nm will more move 14nm productions or possible if cost measure just to keep them on 22nm.

    The big question was Chipset ever on 14nm?
  • III-V - Thursday, December 13, 2018 - link

    Their chip sets have been on 14nm, yes. Thus the interest surrounding this product. It's a little unusual to port a product to an older node.
  • III-V - Thursday, December 13, 2018 - link

    From what I heard, part of the reason why they moved their chipsets to 14nm on such an aggressive schedule is for compliance with California's recent energy regulations.
  • Gondalf - Saturday, December 15, 2018 - link

    Still now they have a new 22nm process named 22FFL that has 14nm power consumption or lower.
    These chipsets are done in 22FFL so they meet or excess the energy regulations.
  • FullmetalTitan - Friday, December 14, 2018 - link

    This pivot is almost certainly to alleviate their line balance issues caused by stagnating at 14nm, we have already seen two rounds of CapEx to expand 14nm capacity, this is just another part of that same effort to shift the load.
  • rahvin - Monday, December 17, 2018 - link

    Exactly. They announced a couple months ago during the 3Q financials they were going to do this to try to alleviate the bottleneck they've got at 14nm now because of the 10nm screwup. They can't make enough CPU's to satisfy demand because of this so they are pulling products out of the lineup and backporting them to make more capacity available for CPU's.

    There will be other products beyond this that get moved as well.
  • eastcoast_pete - Wednesday, December 19, 2018 - link

    Intel being Intel, their marketing department will call it the "classic edition" or "classic node", and charge us extra for the privilege of them using a "mature" process again.

Log in

Don't have an account? Sign up now