GLOBALFOUNDRIES Throws in the Towel on 7nm FinFET Process Development

Discussion in 'HardForum Tech News' started by cageymaru, Aug 27, 2018.

  1. cageymaru

    cageymaru [H]ard as it Gets

    Messages:
    19,814
    Joined:
    Apr 10, 2003
    GLOBALFOUNDRIES (GF) has placed an indefinite hold on the development of a 7nm FinFET program and is instead focusing on 14/12nm FinFET. GF is also reducing their workforce by 5%. This leaves only Intel, TSMC, and Samsung to develop the most cutting edge processes.

    "Lifting the burden of investing at the leading edge will allow GF to make more targeted investments in technologies that really matter to the majority of chip designers in fast-growing markets such as RF, IoT, 5G, industrial and automotive," said Samuel Wang, research vice president at Gartner. "While the leading edge gets most of the headlines, fewer customers can afford the transition to 7nm and finer geometries. 14nm and above technologies will continue to be the important demand driver for the foundry business for many years to come. There is significant room for innovation on these nodes to fuel the next wave of technology."
     
  2. Armenius

    Armenius I Drive Myself to the [H]ospital

    Messages:
    18,618
    Joined:
    Jan 28, 2014
    RIP Vega refresh.

    I'm an idiot. AMD is using TSMC for 7nm.
     
  3. cageymaru

    cageymaru [H]ard as it Gets

    Messages:
    19,814
    Joined:
    Apr 10, 2003
    TSMC for all new process tech.
     
    Nobu and jnemesh like this.
  4. Armenius

    Armenius I Drive Myself to the [H]ospital

    Messages:
    18,618
    Joined:
    Jan 28, 2014
    Beat you to the correction while you were typing :p.
     
    cageymaru and jnemesh like this.
  5. _mockingbird

    _mockingbird Gawd

    Messages:
    992
    Joined:
    Feb 20, 2017
    Hasn't AMD been trying to get out of the Wafer Supply Agreement (WSA) since forever?

    This is a godsend for AMD.
     
    N4CR, nEo717, SickBeast and 2 others like this.
  6. Gideon

    Gideon 2[H]4U

    Messages:
    2,303
    Joined:
    Apr 13, 2006
    Interesting, guess TSMC will be the 7nm will see for AMD.
     
  7. alamox

    alamox Gawd

    Messages:
    596
    Joined:
    Jun 6, 2014
    with the imited supply of 7nm, the products using that not will be priced high, and with GF throwing the towel that makes it even worse.
     
    nEo717 and anthrex like this.
  8. jnemesh

    jnemesh [H]ard|Gawd

    Messages:
    1,084
    Joined:
    Jan 21, 2013
    Says the person who knows nothing of TSMC's actual production capability at 7nm, nor how much GF was expected to produce....
     
    alxlwson and dragonstongue like this.
  9. Oldmodder

    Oldmodder Gawd

    Messages:
    707
    Joined:
    Aug 24, 2018
    I am thinking Global Foundries have gone "smart" and fired just exactly those 5% that actually can and will give a damn, and kept all the brown noses that dont give a damn.
     
    spintroniX and Uvaman2 like this.
  10. pcgeekesq

    pcgeekesq [H]ard|Gawd

    Messages:
    1,399
    Joined:
    Apr 23, 2012
    Whatever that means. How comparable are TSMC's, Samsung's, and Intel's next-gen-geometry processes?
    Or rather, how comparable will they be when they are all actually shipping volume product?

    Who knows?

    We may be heading for a boring period where process improvements are rare and minuscule. It may take an unpredicted breakthrough in one or more of physics, chemistry, optics, and material science to provide the kind of improvement we used to regularly see in every other generation of Intel's CPUs. And that's a little sad.
     
    Ironchef3500 likes this.
  11. Meeho

    Meeho [H]ardness Supreme

    Messages:
    4,470
    Joined:
    Aug 16, 2010
    That was a nice way of saying "we failed".
     
    N4CR, Talyrius, jfreund and 5 others like this.
  12. travm

    travm Limp Gawd

    Messages:
    354
    Joined:
    Feb 26, 2016
    Fake news, GF will launch 7nm Q2 2019, and beat intel and nVidia to death and both companies will be bankrupt by 2020, all AMD chips will destroy everything.
    /s

    What is with tech news, this back and forth shit with half of it clearly just made up is getting old.
    Not saying this story isnt true, but seriously, i've heard so much random shit and quotes from "industry sources" that end up all wrong.

    Maybe i'm just getting old and should read the local livestock prices instead.

    *edit, and then i read the actual article and its GF website themselves LOL. But still, tech news.
    Shoulda bought TSMC stock?
     
  13. Chimpee

    Chimpee [H]ard|Gawd

    Messages:
    1,451
    Joined:
    Jul 6, 2015
    Probably won't affect TSMC stock price much since AMD already committed to them for 7nm production.
     
  14. os2wiz

    os2wiz Gawd

    Messages:
    543
    Joined:
    Nov 20, 2011
    Silicon will be dead within 5 to 6 years for leading edge cpu, gpu, and memory products. Other materials are in development and will be ready before 5 years to advance computing into its next golden age.
     
    dragonstongue and LuxTerra like this.
  15. Qthulu

    Qthulu [H]Lite

    Messages:
    103
    Joined:
    Apr 19, 2018
    Not a big surprise. We all smelled the blood in the water in glofo's 7nm. But this will leave them in a bad spot in 2-3 years.
     
  16. gamerk2

    gamerk2 [H]ard|Gawd

    Messages:
    1,628
    Joined:
    Jul 9, 2012
    We've known that for a while; experts seem to think we can do 4nm on existing tech (which was in doubt for a while), but below that physics itself (transistor leakage) likely becomes an unsolvable problem. There's POSSIBLY one more die shrink; after that, silicon really doesn't have any "free" performance improvements left in it.
     
    juanrga and LuxTerra like this.
  17. gamerk2

    gamerk2 [H]ard|Gawd

    Messages:
    1,628
    Joined:
    Jul 9, 2012
    I read this same exact statement 5 years ago. And 5 years before that.

    Granted, finally reaching the end of die shrinks will help speed the process up a bit, but you're still looking at a totally new architecture/construction. It's going to be a decade or so until you see anything that's ready to replace silicon in the consumer space.

    In short: Get ready for a performance plateau.
     
    juanrga and SickBeast like this.
  18. pcgeekesq

    pcgeekesq [H]ard|Gawd

    Messages:
    1,399
    Joined:
    Apr 23, 2012
    For single-thread, certainly.
    For multi-threaded apps, I expect CPU and GPU vendors to throw down more and more cores.

    And don't be surprised if SMT/HyperThreading goes the way of the dodo: when performance is limited by the watts per square millimeter you can pull off the die, it may not make sense to execute two threads at once on the same tiny area of silicon. Pulling out all the SMT/HT support from the cores will make them smaller and more energy efficient-- and therefore faster-- and probably less vulnerable to side-band attacks too.
     
  19. Lakados

    Lakados [H]ard|Gawd

    Messages:
    1,663
    Joined:
    Feb 3, 2014
    14/12 nm is fine for most everyday electronics and those get ordered in bulk. So I can see them trying to position themselves for that market, they can then transition to smaller processes once the process is refined by other companies And it is less of a guessing game.
     
  20. workshop35

    workshop35 Gawd

    Messages:
    575
    Joined:
    Nov 24, 2013
    Hooray for phase change
     
  21. gamerk2

    gamerk2 [H]ard|Gawd

    Messages:
    1,628
    Joined:
    Jul 9, 2012
    Disagree on both counts.

    You can't just keep adding cores. For one, your limited by die space/yields, Secondly, there are major OS scheduling concerns as core count ramps up, and you saw that in the 2990x benchmarks where scaling between 16 and 32 cores was significantly less then you would expect.

    As for SMT, the reason for SMTs existence is you can get close to an additional core of performance [for certain workloads] for just a fraction of the cost/die space. For example, HTT adds just 10% to the CPU die in exchange for ~30% boost in performance.
     
  22. Uvaman2

    Uvaman2 2[H]4U

    Messages:
    3,124
    Joined:
    Jan 4, 2016
    I think we will see chips the size of my face before silicon is "dead". I know big die chips and mass parallel processing don't make sense now, it will in 10 years or so, when these investments are legacy and silicon goes down in price tremendously.
     
  23. Uvaman2

    Uvaman2 2[H]4U

    Messages:
    3,124
    Joined:
    Jan 4, 2016
    So I take it Intels check cashed?
     
    jnemesh likes this.
  24. Uvaman2

    Uvaman2 2[H]4U

    Messages:
    3,124
    Joined:
    Jan 4, 2016
    I guess i' ve seen this too huh? Its disgraceful.. i guess the problem is outspoken (with logic and merit) often comes with competent, so instead of paying attention to WHAT the person is saying, its only treated as an irritant.
     
  25. Stimpy88

    Stimpy88 [H]ard|Gawd

    Messages:
    1,271
    Joined:
    Feb 18, 2004
    This company has never been as good as the other fabs. But now they publicly state it, the only companies that will use them going forward will be the VIAs of the tech world.
     
  26. DuronBurgerMan

    DuronBurgerMan [H]ard|Gawd

    Messages:
    1,324
    Joined:
    Mar 13, 2017
  27. pcgeekesq

    pcgeekesq [H]ard|Gawd

    Messages:
    1,399
    Joined:
    Apr 23, 2012
    You're assuming perf per square mm of silicon is a relevant metric. I contend it won't be anymore, especially as processes mature and yields rise.
    Perf per watt will be, and not just in portable devices. It's a physics-driven limit: ICs can only run so hot (especially since leakage and electromigration go up with temperature), and there's a limit to how fast you can extract heat from the die without going to expensive cooling technologies (that may be completely inappropriate for a portable device and dense servers).
     
  28. pcgeekesq

    pcgeekesq [H]ard|Gawd

    Messages:
    1,399
    Joined:
    Apr 23, 2012
    They already exist, and have for at least eight years, if your face is about 8x8 inches: Canon develops world's largest CMOS sensor

    "At 202 x 205 mm, the newly developed CMOS sensor is among the largest chips that can be produced from a 12-inch (300 mm) wafer"
     
  29. Galvin

    Galvin 2[H]4U

    Messages:
    2,695
    Joined:
    Jan 22, 2002
    Wasn't die stacking suppose to help with this issue? But so far I've only seen it used in memory chips and flash. Unless it can be used on cpu and gpu's. We may have an issue. Problem is the die size of a gpu is really large. So its much harder to get good yields for gpus.
     
  30. TurboGLH

    TurboGLH Gawd

    Messages:
    563
    Joined:
    Dec 19, 2002
    Based on Juanrgas theory of German buying preferences, we should be seeing a large drop in German sales, now that TSMC will be making almost all of AMDs parts.
     
    N4CR and ccityinstaller like this.
  31. juanrga

    juanrga Pro-Intel / Anti-AMD Just FYI

    Messages:
    2,550
    Joined:
    Feb 22, 2017
    In the best case new materials would eliminate current silicon frequency wall and allow for a new GHZ race. But new materials aren't going to eliminate the ILP wall neither limits imposed by stuff such as Amdahl law. So a new golden age is not coming from going beyond silicon.
     
  32. juanrga

    juanrga Pro-Intel / Anti-AMD Just FYI

    Messages:
    2,550
    Joined:
    Feb 22, 2017
    Apart from limits imposed by Amdahl's law to multithreading (there is a limit beyoind which adding more cores doenst speed up), more cores require more space in the die, so we hit the performance plateau again once reaching silicon shrink limits, as we are doing. And better we don't talk about costs

    Dltjnf-WwAAgmmG.jpg
     
  33. juanrga

    juanrga Pro-Intel / Anti-AMD Just FYI

    Messages:
    2,550
    Joined:
    Feb 22, 2017
    Zen2 was always TSMC. AMD tapeout Zen the past year on TSMC.

    Incorrect conclusion. Didn't you read what I wrote about brand fidelity?

    And a question to you, what Tomsharware sub has officially critized the recent "buy Nvidia" scandal of central Toms? Tomshardware France? Tomshardware Germany? Other?
     
  34. pcgeekesq

    pcgeekesq [H]ard|Gawd

    Messages:
    1,399
    Joined:
    Apr 23, 2012
    Die stacking is used for memory products because heat dissipation is not a major problem there.
    For CPUs, heat dissipation is a primary limiter of performance, and stacking die makes it worse.
     
  35. TurboGLH

    TurboGLH Gawd

    Messages:
    563
    Joined:
    Dec 19, 2002

    I've seen your prattle on about a number of things. I think you covered your position pretty well in this thread.

    https://hardforum.com/threads/intels-desktop-marketshare-in-mindfactory-de.1963497/

    Your claim is that Germany is a country biased towards buying AMD not only because they built and owned fabs there before 2009, but that the appreciation of the German people carried over until today. It's the same nonsense where you tried to use the etymology of the word fan in a tweet by Lisa Su to prove conspiracy theorist style that even AMD was aware of the fanatic way they are viewed in Germany.

    Ignoring that all 12/14nm production was in NY, this should put the final nail into your fucking nonsense about German sales.
     
    schmide likes this.
  36. SighTurtle

    SighTurtle [H]ard|Gawd

    Messages:
    1,412
    Joined:
    Jul 29, 2016
    Wouldn't be surprised if GF had problems with 7nm, but the argument they make with concerns to R&D cost is pretty substantial frankly, as Anandtech's article points out, the owner of GF, Abu Dhabi's sovereign wealth fund Mubadala, has spent over $20 billion on GF without the company making a profit yet. For them to pursue 7nm means, as anandtech points out, One, maintaining future R&D into 5nm and 3nm to remain competitive vs TSMC and Intel, either making a new or upgrade a Fab, and risking current revenue producing volumes now. Thats a lot of freaking money to invest in a company that already struggles to find customers now. Like another $20 billion of investment. So cutting 7nm and just focusing on making money on 14nm and bigger, sounds like a good idea for GF tbh.
     
  37. juanrga

    juanrga Pro-Intel / Anti-AMD Just FYI

    Messages:
    2,550
    Joined:
    Feb 22, 2017
    So you ignored my point and my arguments once again, but now you also ignored my question made to you. The answer was "Tomshardware Germany".
     
  38. TurboGLH

    TurboGLH Gawd

    Messages:
    563
    Joined:
    Dec 19, 2002
    Your argument is nonsense, you've covered it ad nauseum and it still rings with strong notes of simple thoughts tinged with bias. Does every point you make need to be reargued until everyone is tired of listening?

    As far as TH and buy Nvidia,


    Steve at GN also STRONGLY criticized the buy Nvidia..... Scandal (strong word), is he part of the brainless slobbering masses of AMD fans also, or perhaps critcial thinking exists all over the world.
     
    Last edited: Aug 29, 2018
  39. filip

    filip [H]ard|Gawd

    Messages:
    1,792
    Joined:
    Aug 15, 2012
    YES!
    critcial thimkith exists. LOL

    Funny how that made the argument less powerful when it was the punch line. LOL
     
    TurboGLH likes this.
  40. DuronBurgerMan

    DuronBurgerMan [H]ard|Gawd

    Messages:
    1,324
    Joined:
    Mar 13, 2017
    Thought so. Wanted to confirm.