Comments Locked

24 Comments

Back to Article

  • Morawka - Thursday, February 9, 2017 - link

    I wonder if they will use silicon germanium and EUV on their 7nm. i would suspect so since they reserved this site for 7nm. If they were using traditional techniques, they would have probably upgraded existing facilities instead of starting fresh with fab42.
  • Stochastic - Thursday, February 9, 2017 - link

    The generational gains, both in terms of IPC and load power efficiency, have been minimal since 2012 for Intel. Is there any sign of this changing in the next five years? Or are we just too deep into the diminishing returns curve for Intel's architecture?
  • ddriver - Thursday, February 9, 2017 - link

    They mostly care about how small they can make the chip to maximize profit. But it is not entirely their fault, process will soon hit a brick wall and come to a stand still. And all the "graphene and nanotubes will revolutionize chips" stuff is nowhere near commercially feasible, nor is the outlook on new process technology optimistic.
  • Stochastic - Friday, February 10, 2017 - link

    I thought we were past the point where node shrinks decrease per CPU production costs?
  • willis936 - Friday, February 10, 2017 - link

    I can't think of a reason that would ever be the case. If it takes you $10,000 to make a wafer wouldn't you rather get 120 chips out of it rather than 100? There's always an economic benefit to making it smaller. When it becomes too costly to make it smaller then they'll stop going smaller. I don't think performance drives the technology as much as money.
  • 3DoubleD - Friday, February 10, 2017 - link

    I think the cost per chip hasn't been falling significantly or at all because the die sizes haven't been decreasing. In the recent i3 K-series review on Anandtech, they showed a plot that demonstrated this perfectly - die sizes are stagnant. YES, you get more transistors and performance with each generation and the cost/transistor or cost/performance drops, but this is because more transistors are crammed into the same area or are utilized more efficiently.

    With die sizes stagnant, the cost per wafer is stagnant or increasing unless they can increase the wafer size in a profitable way. This comes with its own issues though.

    Factoring in that the technologies required to produce "7 nm" node wafers are temperamental (EUV) or has reduced throughput (multipatterning), price per wafer (and per die at fixed die sizes) is bound to increase (or have already started to increase).
  • ddriver - Friday, February 10, 2017 - link

    It is curious that die sizes remain stagnant while at the same time that is also true for performance. It begs the question, if the transistor count increase is nowhere near the modest performance gains we see, what exactly is intel investing those transistors into?

    Skynet?

    From the looks of it, we might actually begin to see regressions in terms of efficinecy in upcoming process nodes. All in the name of shrinking things down and cramming more transistors in there. But why? Maybe intel is just bent on keeping moore's nonsensical law alive, which in their interest is about transistor count only, not about performance or any meaningful metric, are they merely cramming more transistors for the sake of keeping the pipe dream of infinite scaling alive, or are they doing something pointless, or even sinister with those transistors?
  • phoenix_rizzen - Saturday, February 11, 2017 - link

    Most of the die goes to the GPU these days. The actual CPU cores are tiny in comparison.
  • HStewart - Thursday, December 6, 2018 - link

    First of all this is not just Intel on investing where transistors are going. As for Skynet and AI push, I believe we are seeing that more on GPU side - especially with NVidia and even ARM based stuff.

    As for keeping size down, it probably more aim at mobile - but if you think about - smaller the distance less electrons have to travel.

    As for infinite scaling, Xeon scalable can scale up to 500+ cpus - and that independent of cpu cores. I am not sure this is concern, but I would think that huge dies would be fragile.

    One thing I like is idea of EMiB, it is nice idea that CPU and even GPU can be smaller size chip while IO and other components can be larger size chip on same package. It just seem not that logically to put everything on more dense chip unless you have the ability. Also mix and match of components is a good things - especially when discussing mobile world.

    The world has change, mobile world is pretty much driving the technology and desktop world is just coming a long for the ride.
  • Jaybus - Tuesday, February 14, 2017 - link

    Not really. The process equipment also gets more complicated and more expensive. The Van der Waals radius of Si is 210 pm. 7 nm is about 17 atoms wide. It is just incredibly small. The hard limit is of course one atom, but the reliability of current flow erodes well before that. More than likely, not too many die shrinks remain possible for Si-based wafers. My guess is a move to photonic devices, rather than electronic. In particular, photonic chip-to-chip interfaces could play a large role. For example, a photonic memory bus could be many times faster than on-chip RAM, allowing the elimination of on-chip L2 and L3 cache. That alone would reclaim a huge amount of real estate that could then be used to increase logic gate count without a die shrink.
  • kiers - Thursday, February 9, 2017 - link

    Does Arizona have the GROUNDWATER to support this kind of Fab in the desert?
  • seerak - Friday, February 10, 2017 - link

    How much groundwater does a fab need?

    I don't actually know the answer to that - but that facility is surrounded by several subdivisions with fake lakes, so there's water to be had.
  • GibsonSG - Wednesday, March 8, 2017 - link

    Should have plenty...mainly due to the National Reclamation Act of 1902 which, over time, has seen a chain of lakes created NorthEast of the metro-Phoenix area to supply the city/state with water. Also, to build or expand, the city or private water company has to prove to the state that they have a 100-year assured water supply. The state/city also make extensive use of recharge and reclaim systems. On top of all that, I believe Chandler's Ocotillo campus has on-site recycling for about 80% or more of the water they use.

    Intel would never have started FAB 42 to begin with if they didn't have supply assurances, considering the investment cost.
  • paperfist - Friday, February 10, 2017 - link

    What are those lush green fields to the right of the facility?
  • IntelToaster - Friday, February 10, 2017 - link

    From what I can see on Google Earth, it looks like agricultural flood irrigation plots. I must be mistaken, as I can't see why anyone would want to use flood irrigation in Arizona, but I couldn't find any further information on it.
  • Refuge - Wednesday, February 15, 2017 - link

    Actually fun fact, Arizona is plagued by flash flooding.

    The ground is too dry and compact to absorb any moisture, so even a short thunderstorm turns into a flash flood within minutes.

    Gotta keep dem chips dry!
  • SkyBill40 - Monday, March 13, 2017 - link

    Farm land. There's a lot of corn fields in the area.
  • hrrmph - Friday, February 10, 2017 - link

    Enough butchered English. Get this author an editor.
  • zodiacfml - Saturday, February 11, 2017 - link

    This is going to take a long, while. On the conversation on performance, it is more about competition which caused the stagnation. We should be using larger die size chips than we have these days.
  • karakarga - Sunday, February 12, 2017 - link

    22nm to 10nm fabric cost: 5 billion dollars,
    7nm and lower fabric cost: 7 billion dollars.
    Maybe 3nm and lover fabric cost: 10 billion dollars

    Do we also pay "double", after 5~10 years later? Or more?

    Because more people are leaving desktop computers, going straight for cell phones. Many notebook producers are lovering or completely stopping notebook production, like Sony. Even Samsung unnoticable at shops.

    If it goes like that, earlier than reaching the shriking limit wall, (which seems 1nm) Intel may give up production, because of rising costs! If no profit, no new factory, thus no production.

    May be we are 10~15 years close to doomsday of desktop computers.....
  • HStewart - Thursday, December 6, 2018 - link

    I would say that people are not replacing computers for cell phones - but cell phone are addition and in some cases people don't need a computer and just use cell phone

    Sony and Samsung are both bad examples of this, I don't believe either company has ever made a desktop computer Samsung has be mostly android - but starting with Samsung TabPro S, Samsung has made a windows based tablet.

    Apple is the weird case, not going touch on Mac lines, and pretty much desktop is gone unless you consider mini and all in one desktops

    Intel is not going to give up - technology will change - with recent changes in Intel personal and FABs, I think it is Intel haters dream that Intel will give up production.

    Only thing stated here possibly right, is doomsday of desktop computers but It likely less than 10 years - as soon as mobile processors are as fast as desktop and can cheaply get external GPU's if required for mobile, desktop is gone. For majority of uses it is already, just go into local BestBuy and count Desktop vs Laptop or even include ARM Tablets
  • HStewart - Thursday, December 6, 2018 - link

    But it does cost to make smaller technology, but that is for every company not just Intel
  • HStewart - Thursday, December 6, 2018 - link

    I actually pay double for same level of technology for time period 10 or so year ago.
    Cost will actually go down.

    Example Dell Inspiron 7000 with 300Mhz Pentium 300 cost $3500. My dell XPS 13 2in1 runs circle around it and cost $1100. Ten years ago my Supermicro Dual Xeon 5160 cost me $8000 but until i5/i7 series came out nothing could touch it. Now the same XPS 13 2in1 above can out performed with i7 Y cpu.
  • SkyBill40 - Monday, March 13, 2017 - link

    I live right up the road from the plant and it's quite the footprint. It'll be interesting to see what the future holds for Chandler as well as Intel. Given how AMD has had a resurgence of sorts with the Ryzen launch, Intel is going to need something greater than the refresh it's milked out of the last few series. Will that come on the 7nm fab? I suppose we'll see.

Log in

Don't have an account? Sign up now