Intel's upcoming 10nm and beyond

Discussion in 'Hardware Components and Aftermarket Upgrades' started by ajc9988, Apr 25, 2019.

  1. ajc9988

    ajc9988 Death by a thousand paper cuts

    Reputations:
    1,521
    Messages:
    5,663
    Likes Received:
    7,994
    Trophy Points:
    681
    That is a correct interpretation. By going with smaller chiplets, you increase yields drastically.

    As an example, let's look at Intel's 28-core behemoth on the 14nm++ node. With each "+", Intel decreases density, which allows for higher clocked frequencies. Even with that, Intel's 28-core only achieves around 35% yields. Why? Because by going so large with a monolithic die, you increase the chance of a critical defect occurring on the die, which then makes the chip unusable. Meanwhile, if you make a smaller chip, such as a 4 or 8 core, especially if the architecture allows for shutting off cores completely, allowing for you to make it a 6-core chip, or similar, then you can get actual and effective yields up during the manufacturing process.

    Now, for AMD, and this is only used as a comparison for what Intel is going to adopt, they were achieving in the mid-80s or higher on yields on GF 14nm, a very mature node (same with 12nm). Now, with moving the I/O and IMC off of the core die, if a critical defect hit any of those, the chip would be useless. Even non-critical defects could slow the memory controller's capabilities or you would have reduced PCIe lanes, etc. So by moving those hard to shrink components off, you reduce the R&D costs to reach a new node, reduce the risks of critical and non-critical defects hitting the die, and reduce the die size considerably, allowing for the core die made on the new node to be smaller so each wafer produces both a higher actual and higher effective yield. This is why AMD is rumored with Zen 2 to be achieving a 70% yield on TSMC's 7nm process. That comes up to, if the 70% is actual yield, approximately 506 good dies per wafer, which each wafer for 7nm (or Intel's 10nm) costs around $11,000. That means AMD is yielding around $22.50 per core die on 7nm! That would be 126 32-core CPUs per wafer! Meanwhile, Intel's 35% yield on a mature wafer that costs around $6000 is a problem. That is why Intel's extreme core count CPUs cost so much.

    Now, that doesn't mean Intel's problems are solved by disintegrating the cores into separate dies with chiplets. Intel is rumored to, even with 2 and 4 core chips, have problems with yields on 10nm. They are working on it, but until they show more than 4-core chips, which isn't really even planned for this year as they are primarily using it on mobile "U" and "Y" chips and will have a Xeon server offering that will sit in between the 14nm Cooper lake and the 14nm Cascade lake. This means they are not likely doing a full lineup of 10nm chips for servers, and no plans were shown to suggest Rocket Lake will be on 10nm.

    That, however, does not mean they are not working on accelerating 7nm. By all accounts, they may have 7nm up and running for volume in 2021 (protected forward looking statements under securities law). If they have that, theoretically, they may be able to achieve a higher density than TSMC's 5nm, which will be offered in volume in 2020, but that AMD is not expected to adopt until 2021, while 3nm may not be offered by TSMC until 2023 (unless they successfully move 5nm fabs over to 3nm ahead of the new fab they are building for 3nm). Samsung, however, is expected to offer 3nm in 2021. Intel, TSMC, and Samsung are the last standing cutting edge fabs. IBM sold to GF. GF, after not getting AMD 7nm deal mentioned shopping for a buyer months ago. So, that leaves those big three.

    If Intel cannot get 7nm to go off without a hitch, then there is a real possibility of there becoming just two large fabs: TSMC and Samsung. Hell of a thing to think about, right?

    Now, that is not discussing Intel's architectural engineering, which is awesome. Rather, it is saying that they have major process issues at the moment, which if not resolved, may require them to go fabless.
     
  2. ole!!!

    ole!!! Notebook Prophet

    Reputations:
    2,153
    Messages:
    5,677
    Likes Received:
    3,591
    Trophy Points:
    431
    i think some people just got to admit, AMD went and prepare the smart way of doing things from yrs ago now we are seeing the results while intel is currently doing what AMD was doing 3-4 yrs ago, planning for new arch due to market milking they only got themselves to blame.

    performance is another topic, intel may be able to do 8 cores at 5.5ghz on some water cooling and may still have a decent lead in gaming. i'll take a 12 or 16 cores at 4.6ghz from AMD instead. the frequency temptation gone way down, by having 2x more cores and a lesser price while having similar IPC imo is worth it.
     
    ajc9988 likes this.
  3. tilleroftheearth

    tilleroftheearth Wisdom listens quietly...

    Reputations:
    4,765
    Messages:
    12,217
    Likes Received:
    2,255
    Trophy Points:
    631
    Once again, your assumptions are showing. :)

    If Intel is currently doing this today, they didn't just wake up and said 'lets copy AMD' yesterday. They've been working on it for a very long time too. Agreed that they got a little sidetracked with all that's happened in-between, but they're getting back on track with each passing day. :p

    So, I think that some people just have to admit that the top two will be very close to the other if they want to succeed long term. Nobody is getting game-changing future tech from an alien race yet that they keep hidden in the basement; yeah not even the fruity company could sustain that magical belief for forever... :)

    The hard work still needs to be done and as the best minds keep trading offices between Intel and AMD, the final products will continue to look like copies of the other. ;)

    I will still pick the productivity champion, not a cheaper, but ultimately weaker solution, that just doesn't make business sense or otherwise for a one time cost in return for many years of ROI to be gained though. :)


     
    Papusan likes this.
  4. Talon

    Talon Notebook Virtuoso

    Reputations:
    1,167
    Messages:
    3,202
    Likes Received:
    3,691
    Trophy Points:
    331
  5. ajc9988

    ajc9988 Death by a thousand paper cuts

    Reputations:
    1,521
    Messages:
    5,663
    Likes Received:
    7,994
    Trophy Points:
    681
    I'll believe it when I see it, and it's allegedly Q2.

    Now, I'm wondering since you get 4 working 28-core chips on a mature 14nm++ node, working out to a 35% yield rate, how they will achieve this 26-core part on a process they barely got working for 2 and 4 core low power chips for laptops and mobile?

    As to PCIe 5 and DDR5, Zen 3 Epyc server parts are said to have that and launch in Q1. So that's not special. AMD will have beat them to PCIe for consumers and PCIe for servers, considering Cooper lake is releasing Q1 against the rumored Zen 3. And consumer chips don't get PCIe 4.0 until someone next year.

    So, welcome to the party on something that AMD was rumored to have during the same time frame. Except AMD seems to be making good on their release. Intel's recent record....

    Sent from my SM-G900P using Tapatalk
     
    hmscott likes this.
  6. ole!!!

    ole!!! Notebook Prophet

    Reputations:
    2,153
    Messages:
    5,677
    Likes Received:
    3,591
    Trophy Points:
    431
    yep, assumption for sure, though more likely to be real AND better than anything intel have and will offer in the next yr or two for sure. 10 cores 14nm++ will run damn hot for sho, though 10 cores at 5.1ghz is impressive, im over it because my laptop can't use it, it can barely even do 8 core with normal fan profile.


    @ajc9988 i have little doubt intel wont go full out given how much they know they're behind at least for a yr or two. we know 10nm is definitely coming and server is going to get the best and first of it because its not frequency dependent and unlike mainstream care about more performance, first gen 10nm wont have those high clocks which will be just fine.

    those 10nm yields are low anyway, hopefully they can hit it in the 20-30% LOL but i seriously doubt it especially with 28 cores, on a brand new node that suppose to come out 4 yrs ago.

    but yes this is probably another PR stunt if charlie was to see it to keep the share prices up and give future hope, the similar stunt they pulled with lenovo with 10nm laptop and that was it, nothing else afterwards. they want people ot keep hope and not buy into AMD's product, some will probably sitll be swayed by intel's bs, fanboism runs deep.
     
    ajc9988 likes this.
  7. TANWare

    TANWare Just This Side of Senile, I think. Super Moderator

    Reputations:
    2,500
    Messages:
    9,389
    Likes Received:
    4,789
    Trophy Points:
    431
    You have to act upon known knowledge, not on speculation of the unknown. To do so is irrational.

    Future proof, purchase a CPU and board that can not just drop in an upgraded CPU, that is irrational as well.
     
    ajc9988 likes this.
  8. Talon

    Talon Notebook Virtuoso

    Reputations:
    1,167
    Messages:
    3,202
    Likes Received:
    3,691
    Trophy Points:
    331
    https://www.engadget.com/2019/05/26/intel-10nm-ice-lake-cpu-gaming/

    Intel showing off their 25w TDP Ice-Lake-U with 11th gen graphics. Faster than the latest Ryzen 25w APUs in both CPU and graphics. The Ryzen APU was great for it's graphical performance, but now Intel has single core, multi core, and graphics performance ahead above Ryzen. 10nm shaping up to be promising fast. I am looking forward to an Ultra book from Intel on 10nm soon. :)
     
    tilleroftheearth likes this.
  9. ajc9988

    ajc9988 Death by a thousand paper cuts

    Reputations:
    1,521
    Messages:
    5,663
    Likes Received:
    7,994
    Trophy Points:
    681
    As someone else pointed out:

    "

    Not bad, not bad.

    However, the specs were wildly different in regard to RAM speed:

    * Ice Lake-U LPDDR4X-3733
    * Intel 8565U 16GB DDR4-2400
    * AMD 8GB DDR4-2400 (single channel? really Intel?)

    And also there's a strange mix of low/med settings as if Intel tried to cherry-pick the titles to show the best performance [gains].

    Absolute most consumer laptops won't be running DDR4-3733 RAM which is hugely expensive."

    Sent from my SM-G900P using Tapatalk
     
    Robbo99999 and hmscott like this.
  10. hmscott

    hmscott Notebook Nobel Laureate

    Reputations:
    6,538
    Messages:
    19,856
    Likes Received:
    24,695
    Trophy Points:
    931
    It could be Huawei is counting on Intel failing to deliver a 10nm product on a partner roadmap once again, showing how Intel isn't trustworthy.

    As suggested by the trustworthy Huawei -> that just leaked proprietary NDA'd technical data of a trusted partner. :)

    It is interesting how Intel is showing different roadmaps to different partners, with desktop CPU's left out of some while showing them to others.

    I hope Intel really does deliver 10nm deskop CPU's and laptop H CPU's in production quantities by 2022.
     
    Last edited: May 26, 2019
Loading...

Share This Page