Intel Talks 10nm+ process Ice Lake

Discussion in 'Frontpage news' started by Hilbert Hagedoorn, Aug 16, 2017.

  1. Hilbert Hagedoorn

    Hilbert Hagedoorn Don Vito Corleone Staff Member

    Messages:
    48,324
    Likes Received:
    18,405
    GPU:
    AMD | NVIDIA
    Intel, in it's press-room shared some info on the Coffee Lake generation succesor, Ice Lake. Coffee Lake is fabbed at 14nm, Ice Lake will be fabbed at 10nm. Ice Lake intially will be followed by Can...

    Intel Talks 10nm+ process Ice Lake
     
  2. BlueRay

    BlueRay Guest

    Messages:
    278
    Likes Received:
    77
    GPU:
    EVGA GTX 1070 FTW
    Yes the more cores is what we needed. Shrinking and 2% IPC icrease with few teaks isn't going to cut it. That's why we remained stagnant in CPUs the last 6 years. AMD offered first more cores at a mainstream price and this made all the difference (I do not count Faildozer ofc). Now Intel is following.
    This is the change I was waiting for to upgrade my old 2500K.
     
  3. Matt26LFC

    Matt26LFC Ancient Guru

    Messages:
    3,123
    Likes Received:
    67
    GPU:
    RTX 2080Ti
    I like the fact that they're going straight to their 2nd Gen 10nm for mainstream desktop stuff. Hoping with what AMD has launched this year means that IceLake will come with not only that nice new process but some tasty architectural tweaks too!

    Not sure if IceLake is a brand new Arch or a further refinement of Skylake. But even if it is a new arch I wouldn't expect anything revolutionary from Intel just yet as this is only slated for next year. Perhaps 2019 will be the year for a revolutionary new architecture on 10nm++ followed by a 7nm die shrink of that arch in 2020 who knows

    Hopefully with AMD latest innovations in CPU will mean a more innovation and better products all round.

    Looks like Canonlake will be low wattage parts only and on first gen 10nm
     
  4. Ryrynz

    Ryrynz Active Member

    Messages:
    82
    Likes Received:
    3
    GPU:
    GTX 960
    We've known that for a very long time already..
     

  5. -Tj-

    -Tj- Ancient Guru

    Messages:
    18,095
    Likes Received:
    2,601
    GPU:
    3080TI iChill Black
    Cannon lake is still old SB arch, icelake is new x86, apparently emulating old stuff and new hw with x86 tweaks. Imo the only revolutionary upgrade for anyone looking at new cpu.
     
  6. robintson

    robintson Guest

    Messages:
    423
    Likes Received:
    114
    GPU:
    Asus_Strix2080Ti_OC
    Intel has finally woken up. It was about time, since 2014, nothing really new came up from Intel, regarding CPU's. Now 10 nm CPU's coming in 2018, that is more than four years of sleeping and it's enough. Big THANKS to AMD for Ryzen.
     
  7. Emille

    Emille Guest

    Messages:
    785
    Likes Received:
    27
    GPU:
    1080 Ti Aorus Extreme
    I can't believe people are complaining about die shrinks and product choice. I guess they could always launch one cpu product every 4 years like amd or launch them 15 months late like amd did with vegafail.

    I can't wait for coffeelake. I can keep my ram and get a 6 core cpu that will hit 4.6 on air with a new chipset with 3 m.2 pci 4x slots for under $1000. The same cost as the 7820x on it's own is now which would require a $600 motherboard and new ram kit which are $400 plus for what I would want to get with it.
     
  8. waltc3

    waltc3 Maha Guru

    Messages:
    1,445
    Likes Received:
    562
    GPU:
    AMD 50th Ann 5700XT
    It's kind of funny...but years ago when everyone was on 130nm, I remember Intel doing big PR spreads about 90nm--"coming right up." 90nm hit dram & simple chip production *years* before it hit Intel cpu mass production, IIRC. Right now, Intel is simply struggling to save face, imo. This time around, AMD is not going to hang out a shingle and sit on 14nm cpus and expect to milk them for years into the future in perpetuity. No, AMD will be balls-to-the-walls R&D moving ahead--as the company has already seen the futility of leaping out ahead without plans/means to *stay* ahead. If Intel thought AMD was competitive during the Athlon era, they ain't seen nothin' yet, I would imagine...;)
     
  9. FM57

    FM57 Master Guru

    Messages:
    221
    Likes Received:
    94
    GPU:
    Palit RTX 2070
    10nm seems a big step. We are going to eat and drink 10nm for at least 4 years from 2018. So, no easy way to speed things:
    - the race on gigahertz is over
    - the downsizing is expentially difficult

    Which leaves:
    - cores number explosion (happening)
    - and total change of architecture. Will Intel be bold enough to go that way and leave x86 behind at least for the consumer market ?
     
  10. Stormyandcold

    Stormyandcold Ancient Guru

    Messages:
    5,872
    Likes Received:
    446
    GPU:
    RTX3080ti Founders
    More cores is all well and good.

    However, how many consumers can afford these cpus?

    We'll be waiting a while before 12 and 16 core cpus become mainstream.

    £300 for 8-cores in 2017 is very nice from AMD though.
     

  11. Sixtyfps

    Sixtyfps Master Guru

    Messages:
    242
    Likes Received:
    43
    GPU:
    Tuff 4080 non OC
    If coffee can have a 10%to 15% gain over Kaby I might upgrade my 4790k. Or just ride her out till she dies LOL
     
  12. thatguy91

    thatguy91 Guest

    AMD are looking at their 7 nm Zen 2 around the same time as Ice Lake / Tiger Lake, will be interesting!
     
  13. Backstabak

    Backstabak Master Guru

    Messages:
    860
    Likes Received:
    347
    GPU:
    Gigabyte Rx 5700xt
    Is it really a 10nm, as a gap between the drain and source of the transistor, or is it just the PR name of the more efficient process ? As having more transistors on the same surface could really be cool, but at 10nm i expect that they'll start to see some quantum effects and I'd really like to know how they solved them.
     
  14. ocsystem

    ocsystem Master Guru

    Messages:
    210
    Likes Received:
    13
    GPU:
    AMD 6600xt
    more shrink = less transistors or is the opposite. I still not convinced regarding billions of transistors can fit on so small 32nm or 22nm leave alone 10nm process...
     
  15. Loophole35

    Loophole35 Guest

    Messages:
    9,797
    Likes Received:
    1,161
    GPU:
    EVGA 1080ti SC
    32-22-14-10nm refers to the transistor size not the size of the die area.
     

  16. Neo Cyrus

    Neo Cyrus Ancient Guru

    Messages:
    10,777
    Likes Received:
    1,388
    GPU:
    黃仁勳 stole my 4090
    I'd be shocked if it was anywhere remotely near a 10% IPC gain. Intel haven't done that since they took control in 2006. Even with AMD competing for the first time since then, I don't expect a real increase because these chips were in development before they knew what AMD could really offer.

    You shouldn't need to upgrade anyway, an OC'd 4790K is good for at least another 2 years without bottlenecking any games, I expect longer.
     
    Last edited: Aug 18, 2017
  17. thatguy91

    thatguy91 Guest

    In this case it's not a competition issue, Intel have just had a bad run. It may not seem it, but they have! I've said this a couple of times in older threads, but I'll explain my reasoning again :).

    A separate issue, although still 'important' is Intel not wanting to compete with themselves. The use of cheap TIM material is twofold, the first is because it's cheaper, but it also limits what you can achieve with the chip without modification. This meant their Extreme processors remained 'attractive'. I dare say there are a large number of people who stuffed up a delid, that's a win for Intel as well because the failed attempt meant the people had to buy another Intel processor.

    To the issue, Intel were going quite well until Haswell, and from there the problems started. Broadwell was meant to replace Haswell, but they had serious issues in getting it to work as intented. In the meantime they released Haswell Refresh, an originally unintended update to the Haswell range. Instead of wasting all the R&D time and expenditure into Broadwell they released it as basically a limited run CPU, but no doubt a lot of what they learnt went into Skylake.

    Skylake was a 'success', however, it was intended to be replaced by Cannonlake. As the time approached for the original Cannonlake release it was realised that it simply wasn't going to be ready. Now, either they had planned Kaby Lake as a contigency plan or they pushed it quickly, only Intel knows that, but Kaby Lake was not on their schedule originally. Realistically they should have called it Skylake Refresh, but I guess they didn't want to admit another failure as with Haswell and Haswell Refresh. It's why Kaby Lake isn't much quicker than Skylake.

    Intel were then in the position to replace Kaby Lake with Canonlake, but again delays meant that it wasn't going to be released, so Coffee Lake was born. I suspect Coffee Lake may have been planned as a contigency around the same time they made Kaby Lake, although this has associated cost anything they learnt could go towards future processors. Turns out they could never get Cannonlake to work for high performance scenarios, so they relegated it to low end devices, and the desktop high performance part is Ice Lake, Cannonlake's successor. This also means that on the surface Ice Lake contradicts the now tick-tock-tock cycle of process, but you have to remember it's a second generation 10 nm product from Intel. This is only if Intel can get the 10 nm process right. It may be the case that it will be limited in the processor speed it can achieve, but they'll cover that fact with the likelihood of more cores on the mainstream performance parts. Around the same time as Ice Lake AMD plan to release Zen 2, which is reported to be on the 7 nm process. I think at this point we'll see a huge competition drive between the two, since by then Zen will become an fully established brand and Intel will truly see them as a threat (probably more so than Zen).
     
  18. RealNC

    RealNC Ancient Guru

    Messages:
    4,894
    Likes Received:
    3,168
    GPU:
    RTX 4070 Ti Super
    So Intel will stick to 10nm and larger for a while longer.

    Means there's still some juice left they're not going to squeeze yet. This has been going on since sandy bridge was released. Very, very slow improvements. Milk the market to the maximum.

    I was hoping there'd be a jump to <10nm announced after Ryzen, but it seems Intel does not really care about Ryzen much. Meaning competition isn't working as intended...
     
    Last edited: Aug 19, 2017
  19. qepsilonp

    qepsilonp Guest

    Messages:
    3
    Likes Received:
    0
    GPU:
    GTX 760 2GB
    Has everyone missed that 10nm+ isn't going to increase performance over 14nm++? Like are we looking at different charts... In fact it looks like a small performance regression, and honestly no I don't expect any IPC improvements from Ice Lake, Intel was stuck on 14nm for most of 4 years and didn't think that maybe an architectural improvement necessary? BS until they bring the ground up architecture likely Sapphire Rapids as it is not Ice Lake as Ice Lake has been on the books for the last 2 years, and Intel only started hiring people to work on the ground up architecture last year. (I also expect an IPC uplift with DDR5, but I don't expect that until 2020 and AMD will get the same benefit.)

    While AMD is using TSMC's 7nm for there first Epyc CPUs at least, which yes Intel's 10nm on density when you factor logic and SRAM density is about equal to TSMC's 7nm, but if TSMC's 7nm claims are to be believed it's a 20% speed improvement over there 10nm which was 15% faster than there 16nm which was about 15% faster than Global Foundries 14LPP in measured reality which is what OG Ryzen was made on.

    Like I take those TSMC claims with a grain of salt as those were likely on low power parts so they say 35% I would put it more to 17.5% is more likely, but that 15% over Global Foundries we know that for fact so 37.5% faster is not unreasonable to expect so 5Ghz all core and 5.6Ghz single core is possible, and you wanna want to know what; I am under selling that even more. Because TSMC claims to be able to cut power consumption on 7nm over 10nm by 40% and 35% on 10nm over 16nm at the same time.

    So I am not factoring that and I am cutting there claims on performance in half, like seriously main stream parts with a 5Ghz all core boost clock is being ultra conservative with the numbers we have and Intel on other hand is going to be stuck of 14nm until winter 2019 when AMD is starting to launch there 7nm CPUs for server in Q1 2019, and there 10nm+ CPUs offer no performance improvement over 14nm++, and Intel wont have comparable yields to 14nm++ on 10nm+ either so don't expect more than 8 cores.

    And that brings us to the reports of AMD is going to at least have a 12 cores on mainstream and 48 core server CPU so not only will they have a clock speed advantage they will likely have a 50% core advantage as I suspect Intel will move to 32 cores, not that it will help because Data Centre consumers don't want there 28 core CPUs because in AVX workloads they consume about 334W even data centre customers who can afford the $10,000 per CPU don't want them because there power hogs.

    And the fact that AMD is likely to increase IPC 5% to 7% putting them marginally ahead on IPC and you begin to realise...Intel is F******************************************************KED.

    There is literally no saving grace for Intel apart from inertia and AVX 512, that they could likely improve there's your IPC improvement which you almost certainly wont see because no consumer applications uses AVX 512.

    Intel Is basically dead until 2021 when 7nm and a ground up architecture comes around, END OF STORY!

    Oh sorry I suppose you have the 9900k and 9700k coming they look like nice CPUs, if you need to upgrade now they will be you best choice, they stomp Ryzen. But if you can hang on for at more 10 months from now probably 7 to 9 months from the launch of those CPUs AMD will eviscerate them.

    Although I do believe Intel will steam roll AMD in 2021, not to the same degree probably, but because they wont have lost all there inertia because they have been top dog for 12 years, you don't break that momentum completely within 2 years.

    So I think you will see things quickly return to what they are like now, AMD might surprise me, but I doubt it.
     
    Last edited: Aug 2, 2018
  20. -Tj-

    -Tj- Ancient Guru

    Messages:
    18,095
    Likes Received:
    2,601
    GPU:
    3080TI iChill Black
    Nvm, slow brain xD
     
    airbud7 and StewieTech like this.

Share This Page