Intel 10nm/7nm CPU/GPU Rumor Thread : Cannonlake, Icelake, Tigerlake, Sapphire Rapids, etc.

Discussion in 'Intel Processors' started by Dayman, Oct 13, 2017.

  1. IdiotInCharge

    IdiotInCharge [H]ardForum Junkie

    Messages:
    12,435
    Joined:
    Jun 13, 2003
    If they're backporting the fixes in Ice Lake, then they're likely going to range from ahead to noticeably ahead relative to Zen 2, while maintaining their significant sustained clockspeed advantage.

    The 'patched Skylake' cores might be the sweetspot for most consumer uses; 5.0GHz+, up to ten cores, new IGP block for transcoding and streaming, vulnerabilities addressed to include perfrormance no longer being impacted?

    I'd take it over Zen 2.
     
    Dayaks likes this.
  2. oleNBR

    oleNBR Limp Gawd

    Messages:
    371
    Joined:
    Dec 9, 2016
    actually, the 18% ipc is more so from avx2/512 workloads but there are still improvement nonetheless. I calculated rough IPC gain in CB 15 from icelake vs coffeelake to be around 7-8% which is decent. tigerlake will probably be another improvement around 8% im guesing, i have way less info on tigerlake so it is less concrete where as icelake mobiles are out and there are people in notebook forum have tested CB15.
     
  3. Jandor

    Jandor Limp Gawd

    Messages:
    294
    Joined:
    Dec 30, 2018
    Until 10nm Intel won't make improvements, especially since they patch their optimized unsecured cache system. Some good optimizations are coming with 10nm but can't say when on desktop if ever, and the big IPC jump will come with the new architecture associated with 7nm EUV and that will happen probably in 2022 if we listen to Intel. But I don't completely exclude the fact that they are lying and hide a more agressive agenda, because AMD is clearly not waiting so they can catch up. Mind that Zen 3 will be a little optimization. I bet this is going to be a price drop and some frequency improvements, but Zen 4 is going to be huge and this is going to happen in a year and a half. Wait for double the number of cours and 4 thread SMT cores (not sure 4 threads SMT won't come with Zen 3) and TSMC will provide AMD with 2nd genration EUV technology, and this is 1 year before Intel even starts testing his 7nm EUV in their planning. So I'd rather not believe Intel will be so late to follow TSMC (and Samsung) on EUV technology. So AMD may be already selling for half a year 5nm EUV Zen 5 chips, which we don't know anything about and probably not even AMD yet, when Intel will start some 7nm EUV CPU production. This looks really to awful for Intel. I bet there is another hidden plan.
     
  4. oleNBR

    oleNBR Limp Gawd

    Messages:
    371
    Joined:
    Dec 9, 2016
    tbh i just re-did calculation and results are all over the place. because info i pull from are coming from notebookcheck and obviously they are testing different CPU within different laptop. I try to find the highest score of 25w test but for single threaded performance to ensure no power throttling thus turbo is at it's max for that cpu therefore for much easier comparison. result came out to be around between 8-13% IPC increase from CB15 ST (difference could be due to ram speed as well).
     
    IdiotInCharge likes this.
  5. Nightfire

    Nightfire [H]ard|Gawd

    Messages:
    1,941
    Joined:
    Sep 7, 2017
    IdiotInCharge likes this.
  6. IdiotInCharge

    IdiotInCharge [H]ardForum Junkie

    Messages:
    12,435
    Joined:
    Jun 13, 2003
    Not a bad route to take given their current predicament really
     
    Nightfire likes this.
  7. Jandor

    Jandor Limp Gawd

    Messages:
    294
    Joined:
    Dec 30, 2018
    Big big news.

    Intel Canada just announced Intel will be back with 10nm Desktop CPU at the very beginning of 2020.
    No information yet with the core count, but the guy also said the the production problems Intel experienced was because they are massively converting their 14nm facilities into 10nm. There are already 2 facilities producing 10nm and a third one is just starting and the rest to follow, while 7nm and 5nm are perfectly on track with one facility already able to produce 7nm which will deliver in, 2021. 10nm has achieved very good yields.
    All the news posted about new 14nm CPU on schedule at 14nm where fake or simple give away stocks produced at the beginning of this year and renamed into new series to be more attractive.
    So don't upgrade anything for AMD yet until everything about Intel 10nm is officially announced and that shouldn't take too long. People at Intel Canada aren't going to leak if they weren't supported by the head of Intel.
    https://www.itworldcanada.com/artic...-desktop-set-to-arrive-early-next-year/423481
     
  8. Jandor

    Jandor Limp Gawd

    Messages:
    294
    Joined:
    Dec 30, 2018
    An, holly shit, those 12 core Intel Ice lake at 5.2Ghz with much better IPC than Zen 2, will obliterate the 16 core 3950X.
     
  9. Jandor

    Jandor Limp Gawd

    Messages:
    294
    Joined:
    Dec 30, 2018
    It has been tested and the IPC of Ice lake is about 20% better at the same speed than Zen 2.
     
  10. Nightfire

    Nightfire [H]ard|Gawd

    Messages:
    1,941
    Joined:
    Sep 7, 2017
    Wow, your love for Intel would make Juangra blush.

    1. Since your breaking big big news, it is now said they don't really know when it will be released.

    2. Desktop could mean a lot of things ie nuc only

    3. Ice Lake will not hit 5.2 ghz in its wildest dreams.
     
    oleNBR likes this.
  11. Gideon

    Gideon 2[H]4U

    Messages:
    2,365
    Joined:
    Apr 13, 2006
    Ahh there you go destroying all his hope (y)
     
    Nightfire likes this.
  12. oleNBR

    oleNBR Limp Gawd

    Messages:
    371
    Joined:
    Dec 9, 2016
    blind delusion will never get destroyed. the fact that he brings up 5.2ghz already shows you he is delusional, or fanboi at it's max.
     
  13. Nightfire

    Nightfire [H]ard|Gawd

    Messages:
    1,941
    Joined:
    Sep 7, 2017
    https://www.techpowerup.com/261611/...es-in-core-counts-up-to-8-gen12-igpu-included

    4kjyyDSYSvAmujEc.jpg

    So, whereas 2020 Intel laptop buyers have the choice of 6/12 CML with gen 9 graphics or 4/8 ICL with gen 11 graphics with 48 EUs (15w), 2021 buyers are looking to choose between 6/12 RKL with gen 12 with 32 EUs or 4/8 TGL with gen 12 graphics with 96 EUs (28w).

    Really? Even with Tiger Lake (10nm++?), they still can't even get past 4 cores? ICL sort of makes sense for some mobile users over CML as it does well for casual/esport gaming. Even then, they would be better off with a 4/8 CML with a cheapo 1650 card. This will be even more so true with TGL vs RKL.

    This just solidifies that 10nm still can't be packaged for desktop use, even with Tiger Lake.

    As for Rocket Lake? Well, it better have a damn good clock boost or it will be DOA. It will need this for buyers to sacrifice the extra 2 cores of CML and the iGPU improvements will be a non factor as the next AMD APU will cover that front.
     
  14. IdiotInCharge

    IdiotInCharge [H]ardForum Junkie

    Messages:
    12,435
    Joined:
    Jun 13, 2003
    For ultrabooks?

    I was skeptical with their choice to move up to quad core 15w CPUs. They delivered, and realistically you're not going to get more performance out of six or more cores in that form factor. If you load up all six or whatever cores at 15w, you're just going to hit thermal throttling even sooner, so not only do you not gain performance, but you lose performance for more linear workloads that cannot be easily multithreaded.
     
  15. Nightfire

    Nightfire [H]ard|Gawd

    Messages:
    1,941
    Joined:
    Sep 7, 2017
    I wasn't expecting 6 cores in Ultrabooks (Y-config?), but you would hope that they could get 6 cores in a full frame.

    Even focusing on Ultrabooks, the clock differences for the 4/8 Comet Lake and 4/8 Ice Lake Y-series are rather significant.
     
  16. IdiotInCharge

    IdiotInCharge [H]ardForum Junkie

    Messages:
    12,435
    Joined:
    Jun 13, 2003
    Above ultrabooks, Intel already has eight-core CPUs that AMD will very likely struggle to fully compete with. Even if they can get within striking range of performance, mobility is still extremely likely to suffer.

    Which is unfortunate, but despite being stuck on 14nm for most of their lineup to include the SKUs for larger laptops, Intel hasn't been resting on their laurels.
     
  17. Ruined

    Ruined n00b

    Messages:
    17
    Joined:
    Dec 1, 2019
    If it features a backport of "cove" architecture instead of Skylake again then 8c rocket lake will likely outperform 10c comet lake both single and multi. And that would then make sense the reduction of cores.
     
    Last edited: Dec 2, 2019
  18. Nightfire

    Nightfire [H]ard|Gawd

    Messages:
    1,941
    Joined:
    Sep 7, 2017
    I always thought RKL was going to be Sunny Cove as stated in my August post. The last few rumors of it being Skylake never made a lick of sense. Seems that my original instinct could still be correct:
    https://www.techpowerup.com/261735/intel-rocket-lake-an-adaptation-of-willow-cove-cpu-cores-on-14nm

    If this is true, RKL would be wonderful. I know it sounds backwards, but I would rather have RKL than ICL. RKL would give the best of all worlds: Core count, IPC, and frequency.
     
    IdiotInCharge likes this.
  19. IdiotInCharge

    IdiotInCharge [H]ardForum Junkie

    Messages:
    12,435
    Joined:
    Jun 13, 2003
    I'd actually be fine with a *Cove architecture on 14nm. That's all I really wanted out of 10nm; more performance. I don't need cores, really, as much as others claim to, but if I can get more performance out of eight?

    Sign me up.
     
    Nightfire likes this.
  20. Ruined

    Ruined n00b

    Messages:
    17
    Joined:
    Dec 1, 2019
    The other nice thing is it will probably be socket compatible with Comet Lake so those who buy CML early next year will have an upgrade path.

    If the 8c 14nm RKL top end part is true, then this basically confirms sunny cove backport on rocket lake as there would be no other way for Intel to increase perf to get 8c to match 10c comet lake perf on 14nm.

    So we will probably see the below at the top of mainstream:

    Early 2020 - comet lake 10c 14nm skylake refresh
    Early 2021 - rocket lake 8c 14nm sunny cove backport
    Late 2021/Early 2022 (asap) - 7nm big guns

    May see these parts coupled with 10nm in the midrange or low end desktop (like they did with 2019 laptop) but it looks like they can't clock 10nm high enough to be a suitable high end mainstream desktop part process
     
    Last edited: Dec 2, 2019
  21. Nightfire

    Nightfire [H]ard|Gawd

    Messages:
    1,941
    Joined:
    Sep 7, 2017
    Ah good point. CML would be worth it for those getting 4/4, 4/8 or 6/12 core setups to hold them over until the can upgrade to 6/12 or 8/16 RKL rigs.
     
    Ruined likes this.
  22. Ruined

    Ruined n00b

    Messages:
    17
    Joined:
    Dec 1, 2019
    I have to wonder how HEDT will be affected

    For those on a budget that want the pcie lanes flexibility of HEDT with a processor that doesn't suck at games/light stuff and is also good at multicore, the new 10940X actually looks like a pretty good value... But will late 2020 HEDT have cove backport on new Mobo platform that makes this year offerings look like bad value? Hard to say as HEDT usually is behind mainstream in architecture, not the other way around - though Intel is in a unique position now so that may change. They might just put multiple dies per chip to reach 32c-36c but that would make tdp crazy IMO (would remind me of presler).
     
    Last edited: Dec 2, 2019
  23. Snowdog

    Snowdog [H]ardForum Junkie

    Messages:
    9,705
    Joined:
    Apr 22, 2006
    I guess that's better than another Skylake derivative. But recognize that will be a 2021 part, and Intel will still have it's main desktop parts on 14nm. AMD will be on 5nm by then. TSMC high volume 5nm starts next year.

    That is a massive process hole that will be in.
     
    N4CR likes this.
  24. Nightfire

    Nightfire [H]ard|Gawd

    Messages:
    1,941
    Joined:
    Sep 7, 2017
    Not really. 14nm is just an arbitrary number if they can clock higher than 10nm. Advanced 14nm plus Sunny Cove should still have great efficiency.
     
  25. Ruined

    Ruined n00b

    Messages:
    17
    Joined:
    Dec 1, 2019
    Yeah intel 10nm proves that smaller process doesn't always translate to better performance - and that could happen for AMD, too.
     
    Nightfire likes this.
  26. Snowdog

    Snowdog [H]ardForum Junkie

    Messages:
    9,705
    Joined:
    Apr 22, 2006
    That seems like wishful thinking on Intels behalf. AMD has improved performance and lowered power on 7nm.
     
    N4CR likes this.
  27. Ruined

    Ruined n00b

    Messages:
    17
    Joined:
    Dec 1, 2019
    Intel improved performance and lowered power on 14nm. That doesnt mean they did the same on 10nm. Same could happen for AMD 7nm to 5nm, might get improved power but not performance. We will have to see how it shakes out.
     
  28. Mega6

    Mega6 2[H]4U

    Messages:
    2,236
    Joined:
    Aug 13, 2017
    Zen3 new arch on 5nm rumored +15% IPC 4th Q 2020
     
  29. Ruined

    Ruined n00b

    Messages:
    17
    Joined:
    Dec 1, 2019
  30. Mega6

    Mega6 2[H]4U

    Messages:
    2,236
    Joined:
    Aug 13, 2017
  31. Ruined

    Ruined n00b

    Messages:
    17
    Joined:
    Dec 1, 2019
    You are missing the point. IPC increase does not mean that the process will be an overall performance boost. Intel had a 20% IPC increase with Icelake but could not clock it high enough to beat lower IPC Skylake so they stuck with lower IPC 14nm at higher clocks. Hence IPC by itself doesn't mean much. Until real 5nm AMD parts come out there is no way to know how good or bad they will be.
     
    IdiotInCharge likes this.
  32. Mega6

    Mega6 2[H]4U

    Messages:
    2,236
    Joined:
    Aug 13, 2017
    If recent history is worth anything, I have put all my money on Amd. Intel is the one that need to prove itself. How the worm has turned.
     
    N4CR likes this.
  33. IdiotInCharge

    IdiotInCharge [H]ardForum Junkie

    Messages:
    12,435
    Joined:
    Jun 13, 2003
    Well, you would.

    But historically, Intel would be the one to bet on here. If AMD's 15% is actually 15%, then they'll catch up to Ice Lake, just as Intel moves on to faster cores.
     
  34. Mega6

    Mega6 2[H]4U

    Messages:
    2,236
    Joined:
    Aug 13, 2017
    Since tsmc also has the world’s premier fabs. Yeah.
     
  35. IdiotInCharge

    IdiotInCharge [H]ardForum Junkie

    Messages:
    12,435
    Joined:
    Jun 13, 2003
    They got lucky at node shrink bingo.

    Up until their 7nm process, they've either been behind, or been really behind.
     
  36. Mega6

    Mega6 2[H]4U

    Messages:
    2,236
    Joined:
    Aug 13, 2017
    Luck. LoL
    Sure.
     
    N4CR likes this.
  37. IdiotInCharge

    IdiotInCharge [H]ardForum Junkie

    Messages:
    12,435
    Joined:
    Jun 13, 2003
    They picked a line of research and development that paid off. Intel, on a very rare occasion, didn't.

    You're aware that most research involves finding out what doesn't work, right?
     
  38. Mega6

    Mega6 2[H]4U

    Messages:
    2,236
    Joined:
    Aug 13, 2017
    You’re aware that choosing the the correct path is the intelligent, right? Guess intel is researching really hard. Oh wait, they spent more on stock buybacks than R&D. You are aware of that, right?
     
    N4CR likes this.
  39. Ruined

    Ruined n00b

    Messages:
    17
    Joined:
    Dec 1, 2019
    This would be a good observation if Intel hadn't stomped AMD 90% of the time over the past 3 decades. You are in the 10% right now.
     
  40. Mega6

    Mega6 2[H]4U

    Messages:
    2,236
    Joined:
    Aug 13, 2017
    STOMP

    heh heh