Comments Locked

33 Comments

Back to Article

  • shabby - Wednesday, March 17, 2021 - link

    Rocket Lake+ coming next year.
  • Oxford Guy - Thursday, March 18, 2021 - link

    Double + good.
  • JayNor - Wednesday, March 17, 2021 - link

    They already stated in the q4 cc that the 6 month 7nm experiment fixed the yield issue that was delaying them.

    Looks like 10SF and 10ESF are doing better than expected, so no reason to rush into high volume server CPU products on 7nm in 2022.

    I expect them to announce that some GPUs will be done on 7nm in 2022. They already said this was an option for the Xe-HPC compute chiplets.

    Perhaps they will also announce the plan for Grand Ridge. That seems like a good candidate build on 7nm as a first CPU. One of its first appearances on leaked roadmaps did show it as 7nm.
  • Matthias B V - Thursday, March 18, 2021 - link

    Since GPUs are not so high volume yet I expect they keep them at TSMC 6nm for the next years and even move them to 5/3nm as TSMC is dong fine on GPUs. CPUs I rather expect to stay inhouse. Especially since they are a massive volume compared to Intel GPUs the next years.

    I rather expect volume parts (RaptorLakr and SaphireRapids) on 10nm in 2022 and some 5-35 Watt Mobile CPUs (MeteorLake) on 7nm EUV similar to what they did with IceLake 10nm / Comet Lake 14nm.

    And then move to 7nm for the whole protfolio in 2023 depending if they have enough EUV capacity.

    If 7nm is fine on metrics but yield ist just bad they could do so and just eat th elower margin but overall ok capacity (10+7nm). If metrics are not good even that is not an option. That would be really bad for Intel and consumers.
  • jjjag - Thursday, March 18, 2021 - link

    First off, the first Intel 7nm offering will be a PC SOC. Quote from Swan "We now expect to see initial production shipments of our first Intel-based 7nm product, a client CPU, late in 2022 or early 2023". Since Intel has a history of underestimating their issues, add at least 2 Q to that estimate. if Pat G. does not come in with a realistic estimate of LATE 2023, then don't believe anything he says. Swan was CEO until a few weeks ago so nothing has changed yet.

    Secondly, it has nothing to do with "how well 10nm is doing". Intel is way behind in manuf. tech. and they are desperate to catch up. You better believe that if 7nm was healthy they would be in production TODAY. 7nm mostly uses the same mask/litho technology and same fabs as 10nm so you are not taking a depreciation hit by not using 10nm. They WILL "rush into high volume" as soon as they damn well can.

    GrandRidge is a don't care, it's a low volume product for a small set of customers. Raja K. already previewed the new GPU product, "Arctic Sound" and it's on 10nm and possibly external foundaries for the various dies that it uses. They will not even talk about 7nm GPUs until this one gets closer to production.
  • Gondalf - Thursday, March 18, 2021 - link

    Umm yes, 10ESF gives a 15% power reduction advandage over 10SF, so more or less they already have a 5nm equivalent process in power draw but with a larger silicon footprint. Intel have enough wafers to deal with it. Agreed the rush to 7nm is useless in 2021 and 2022.
    The production with EUV right now is a nightmare for TSMC, bad yields, so so pellicles, huge shortages. Not good at all.
  • Silver5urfer - Wednesday, March 17, 2021 - link

    We do not want words in the sky. We need proof to trust Intel.

    They should first address the Fabrication failures and then create a successor to the Ring Bus. Until they can do them AMD is going to chew them apart. Milan will crush Xeon Ice Lake SP since Intel is glued and using 10nmSF vs the tested TSMC 7nm which AMD got a lot of experience from. In HEDT Threadripper is uncontested champion. In Desktop maybe in games Rocketlake might get a few points but in SMT and Productive workloads 5950X chews Intel apart. With more core scaling programs and code, the Intel CPUs *might* be outdated.

    On top shedding the real money maker like 5G to Apple and then getting booted off from Apple, killing 3D NAND market (Intel 480 series SSDs and SLC were pretty dope) and now Optane 3D Xpoint also dead. This is not shedding, damn it they have to first shed the useless crap like Mobileeye and other junk like trying to come into GPU market. Focus on the bread and butter - CPU x86 innovation.
  • domih - Wednesday, March 17, 2021 - link

    Yes (Ryzen and Threadripper user here) but (a) INTEL is going to sell a lot of 11th gen via the usual distributors/integrators (b) INTEL has a massive momentum and large financial assets so they have the advantage for the attrition war with AMD and ARM (c) INTEL can continue to fool around for several years before eventually coming up with "the" new "killer" architecture. The story is ONLY about how many % of the market INTEL is going to lose to AMD and ARM before recovering. Anyway, competition is good. As soon as the supply chains, production and distribution channels recover from the direct and indirect effects of COVID-19, the price war will be bloody. For now, it is not a good time to buy or upgrade hardware, the prices are too crazy.
  • JayNor - Wednesday, March 17, 2021 - link

    Ice Lake Server is not 10SF. That is the Tiger Lake process.
  • JayNor - Wednesday, March 17, 2021 - link

    Intel announce there is no change in their Optane plans.
  • JayNor - Wednesday, March 17, 2021 - link

    Intel only dropped out of 5G smartmodem market. They are in 5G market in a big way with P5900, eASIC, FPGAs, XeonD.
  • GeoffreyA - Thursday, March 18, 2021 - link

    "Focus on the bread and butter - CPU x86 innovation"

    Agreed. I believe part of Intel's failure comes from trying too many things (often leading to dead ends). AMD is winning, partly, because of focusing relentlessly on just one or two things. A single core, which they're pouring all their effort into, and then multiplying it almost like bricklayers. Doing one thing, and doing it well, will always beat being a jack of all trades.
  • GeoffreyA - Thursday, March 18, 2021 - link

    Having said that, Intel Haifa has always been focused on microarchitecture.
  • jjjag - Thursday, March 18, 2021 - link

    Wrong Wrong Wrong Wrong. Intel's failure (not AMDs success) is a combination of hubris and complacency. Hubris - look at the industry papers back ca. 2015. They pushed the density of 10nm TOO FAR, and introduced too many new manuf. features. "Superfin" is really them backing off the density so they can actually produce it now. TSMC has actual (fabless-semi) customers that they pay attention to, so they developed a 7nm/5nm/3nm process that everybody could use and that they could get into production early. Yes they are not as dense as Intel's offerings on the same "node", but it doesn't matter if Intel's volumes are zero.

    Complacency in that they have not improved the micro-architecture of their big core line since Sandy Bridge. Here again they were relying on manuf. to improve performance so they never did anything about the architecture. Also they continually underestimate the competitive landscape and the data that clearly shows ARM cores surpassing X86 performance at low power (where it matters) was dismissed.

    AMD will eventually fail for the same reason, unless they invest more in ARM. ARM is the real future. WIndows is finally getting ported to ARM for real now. The Mac is just the first pebble to roll. Look for boulders to start rolling by 2024/5 with ARM PCs from Dell, HP, Lenovo, Asus, etc.
  • Otritus - Thursday, March 18, 2021 - link

    Actually Intel has improved the micro-architecture of their big cores. Haswell added avx-2 and Skylake is about 25% faster per clock than Sandybridge. Sunny Cove added avx-512 and is about 45% faster per clock than Sandybridge.

    The problem isn't that architectural improvements from Sandybridge to Skylake were small (they were). The problem is that Intel failed in their real bread and butter (in-house advanced mass manufacturing of processors). Sunny Cove would have originally been mass produced in 2017 had Intel not failed with 10nm. Golden Cove would have been on 7nm in either 2019 or 2020 (and this 7nm would be equivalent to TSMC 5nm).
  • Matthias B V - Thursday, March 18, 2021 - link

    Intel has improved but couldn't use as they were all designed for 10nm which was not ready. But I agree there is progress currently and in future.

    AV2 to AVX512 however isn't -it is a waste outside servers and even there it is mixed. AMX will be a little better. They should have something like SVE (Scalable Vector Extension) in x86. What could help x86 extend its lifespan is getting rid of legacy and create a "lean"86. Maybe only 64-Bit and new scalable extentions that can emulate old ones if necessary.

    Overall people want open and flexible archs and whoever delivers will gain. Would also benefit customers as you could see what Intel and a lack of AMD competition did. Overall after losing Apple to ARM it will be harder for x86 as ARM already has momentum and Windows will follow. However this could change if Nvidia really is allowed to buy ARM and kill its independence. In the long run this might turn people to RISC-V.
  • GeoffreyA - Thursday, March 18, 2021 - link

    jjjag, I do believe complacency played a big part. This is what I wrote the other day, replying to a comment on the Rocket Lake article.

    "My guess would be: complacency, underestimating the enemy, putting eggs in too many dead-end baskets, and management that made a mess of excellent engineering talent."
  • Silver5urfer - Thursday, March 18, 2021 - link

    I always wonder what does people have with ARM. For x86 I always see them are more open and easy to build rather than a gated bullshit.

    Android AOSP needs Qcomm CAF and other source to compile proper builds and Kernel HAL drivers if they are not made properly the OS cannot work properly. Same for every single ARM machine. OMAP was the only one which was very open on that front. Qcomm was the reason why SD805 was killed prematurely. Next Graviton2 that's only made for Amazon, AWS. It won't be shipped to anyone else on the market, so let's move from there and it's not going to succeed AMD. Graviton3 ? yeah but by that time EPYC from Zen4 will show up. Qcomm Centriq - Abandoned fully prematurely and saw no big contracts or customers, Altera they recently released the new chip, compared with EPYC no dice. And on top the Software code which is the fundamental basis must be rewritten and optimized. A big no for majority of the enterprises except M$ / Amazon since both are too greedy and want money so save cost on x86 and use ARM. Fujitsu's ARM64 is full custom and no option for outside. Nvidia Automotive processors are ARM maybe that's one place where it can scale, due to limited Power and performance requirements.

    Now consumer devices, Apple is first so lets look at M1, first ARM design based off A14. And cannot beat a Renoir based Zen 2 APU in SMT. ST Spec score and all those first party software makers are being paid by Apple big time, thus allowing for tighter integration. The Mac OS is also downgraded a lot to mimic Tablet bullshit and dumb users, this is not at all a point worth in this discussion but just to note, also Windows10 same ideology killing Win32 for UWP and exerting more control. Corporations want more power since Oil is no more, it's all Tech. And then you have 8cx based Windows machines, full custom Chips and same for x86 AMD SQ1 Surface processors, the reason I boiught these because read "full custom" they do not allow any sort of Bootloader aspects such as Linux and open source OS with proper user control, so any ARM shit is full custom You have maybe Raspberry PI systems which leverage huge DIY thing but it's not competing with commercial proper enterprise grade x86.

    So take your ARM bs somewhere else. AMD will innovate and Intel will be forced to innovate or perish. x86 puts power in people's hand even if Software is castrated (Windows10 to Linux) but ARM takes it all away for some custom BS. SPEC scores ? Muh Apple is world's best CPU ? Go to youtube and click any video showcasing real-world iPhone vs Android flagship comparisons that's all.
  • Matthias B V - Thursday, March 18, 2021 - link

    I wouldn't boy a ARM notebook yet - for now!

    M1 has compeitive performance and amazing battery but it is highly specific and has onboard RAM so unfair to compare. However I expect Intel / AMD to use 3D stacked memory in future as it is enegry efficient and fast.

    Windows on ARM is still crap and QComm SQ1 is barely customized and old. However I expect them to bring customized APUs by 2023 on TSMC or Samsung 3nm. And by then Widndows is ARM ready. I wouldn't underestimate it. Especially since lots of consumers are happy with Office, Netflix, Instagram and Facebook. Battery life will be ridiculous! The only thing that can kill ARM right now is Nvidia takeover.

    Intel is doing a nice move with BigBigger (BigLittle) design but not sure if it is too less too late. Especially if they do not get their 7nm and 5nm running asap. They also should get rid of legacy to safe space and code.
  • Oxford Guy - Thursday, March 18, 2021 - link

    'Apple is first so lets look at M1, first ARM design based off A14. And cannot beat a Renoir based Zen 2 APU in SMT. ST Spec score and all those first party software makers are being paid by Apple big time, thus allowing for tighter integration. The Mac OS is also downgraded a lot to mimic Tablet bullshit and dumb users'

    Much less intelligent users than the people who made the OS that shuts off when you're in the middle of you work a vast success. Microsoft really drew a new line in the psychological sand when it came up with that behavior. 'How much abuse will customers not only put up with but choose to pay for? Let's see!' It's the tech equivalent of Milgram.

    Consumer stupidity (and, especially, passivity) is the norm. It's basically true of tech in general. Otherwise, all the vaunted smart users would have been demanding a true level of competition in all sorts of tech markets (search, GPUs, foundries, ISPs, desktop OS, etc.). Instead, these geniuses spend their time propping up whatever company is the most monopolistic of the quasi-monopolies around. They also blindly throw their money at massive expanding conglomerates when those conglomerates add some 'competition' to a particular market space by entering it — even though enriching them more reduces the competitiveness of tech overall.

    As for your dismissal of M1, it sounds like more fanboy and less objective appraisal. I will say, though, that I find it the height of hilarity that a company with as many billions as Apple has can't manage to make a text editor that doesn't: A) crash multiple times, causing your work to be lost and B) refuse to let you save your open documents after it loads them post-crash.

    Text editing is a brand-new thing in microcomputing, after all!
  • tomatotree - Thursday, March 18, 2021 - link

    Intel has had a successor to the ring bus since 2017, and it was used on Skylake-SP: https://www.anandtech.com/show/11550/the-intel-sky...
  • Oxford Guy - Thursday, March 18, 2021 - link

    'In Desktop maybe in games Rocketlake might get a few points'

    Not at the same power consumption.
  • dsplover - Thursday, March 18, 2021 - link

    I wish them luck. They’ll need it as AMD is stealing more market share than anyone foresaw.
  • Qasar - Thursday, March 18, 2021 - link

    not according to the pro intel fans
  • shabby - Thursday, March 18, 2021 - link

    Microcode isn't final!
  • Qasar - Sunday, March 21, 2021 - link

    doesnt really matter if it isnt final, it wont improve performance all that much, as the follow up article showed, it kind of brought tiger lake up to match 10th gen, but even then still looses to it in some tests, and still, over all, isnt faster then zen3. sorry shabby, but tiger lake, still looks to be a dud.
  • Amandtec - Thursday, March 18, 2021 - link

    Intel may just be screwed because even if they move to TSMC, CISC chip development is known to be more difficult and costly. That RISC would eventually overtake CISC was predicted decades ago - I remember it being discussed in the 90's - now that AMD has the momentum, easier RISC development may carry them further and further away on IPC through regular iterative improvements.
  • kgardas - Thursday, March 18, 2021 - link

    Wake up! Intel is doing RISC business since Pentium Pro in late '90s (and btw, AMD even early with their K5). Internally all their high-perf CPUs are high perf RISC CPUs. That they do not advertise internal ISA and that they translate venerable x86/amd64 ISA to it is another matter.
  • GeoffreyA - Thursday, March 18, 2021 - link

    The main bottleneck in x86 CPUs are the variable-length instructions, making parallel decoding difficult and eating a lot of power. The micro-op cache in Sandy Bridge and Zen has mitigated this to some extent; and marking instructions boundaries in cache---I believe K7 to Bulldozer, and Tremont---can help further. ARM has the advantage of fixed-length instructions, making it easy to work out where instructions begin, and as a result, one sees quite wide front ends on M1, for example.
  • Matthias B V - Thursday, March 18, 2021 - link

    Hope we get lots of information on process and archs. I doubt he will talk about 5nm GAA but focus on 10nm+++ and 7nm.

    Still hope he is less of a bean counter and will go 7nm in 2022. Even if it is parallel to 10nm+++ / Enhanced Super Fin. They need to show they can and get expeience with EUV. Capacity and yield might be low but margin is secondary in my opinion. They can have a 2022 10nm RapotorLake from 5-125 Watt range for desktop and mobile and still sell 7nm MeteorLake in 5-35 Watt Mobile. They did the same with IceLake 10 / CometLake 14nm. And then progress fast to 5nm GAA same shared way in 2024. I do care less about densiy scaling than electical characteristics of GAA. By then they also would have enough EUV capacity which I dount they have in 2022. Not sure if they do as currently anything sells but that would be a short minded strategy if not.

    Also since they cant use their equioment from 22/14/10 nm DUV for 7nm EUV they could move most / some to 10nm and keep the rest to supply Automotive / IOT and get a additional customer base. Those processes are enough or that. Rather build new fabs for EUV and diversify.
  • PaulHoule - Thursday, March 18, 2021 - link

    You'll know that Intel has turned it around when they get their sense of humor back and start naming chips things like: "Death Valley", "Shit Creek", "Love Canal" and such.
  • dsplover - Thursday, March 18, 2021 - link

    How exciting.

    I guess President Biden announcing a “press briefing” 9 days in advance sounded good to Intel too.
  • Micyoliu - Monday, March 21, 2022 - link

    We at https://bestcasinosonlines.com/woocasino/ have the best casinos for you to have fun, play and keep your privacy.

Log in

Don't have an account? Sign up now