news reports broadwell e slips to 2016 but skylake s sampling already
News

Reports: Broadwell-E slips to 2016, but Skylake-S sampling already

Haswell-E is barely out the door, and already, the rumor mill is rumbling about a delay to its successor. According to VR-Zone Chinese, Broadwell-E won't begin sampling until the second quarter of 2015. The site says mass production is scheduled for the first quarter of 2016, which represents a slip from the timeline shown on at least one leaked roadmap from earlier this year. Google's translation doesn't mention anything about VR-Zone's source, but the story does include an official-looking block diagram.

The leaked image suggests Broadwell-E will be similar to its forebear, with "up to" eight CPU cores, 40 PCIe lanes, and 20MB of cache. Support for quad-channel DDR4-2400 memory is also listed—but with only one DIMM per channel. Hmmm.

Although Broadwell-E will be fabbed on a newer 14-nm process, it will reportedly have the same 140W power envelope as Haswell-E. Given the relatively smooth transition from Sandy Bridge-E to Ivy Bridge-E on X79 motherboards—and the Broadwell compatibility boasted by Z97 boards—there's good reason to believe the claim that Broadwell-E will work in existing X99 mobos. Ideally, support will require no more than a firmware update.

In a separate VR-Zone story, "upstream sources" say Skylake-S processors are now sampling to Intel's partners. Those sources provide a picture of one of the LGA1151 chips, which has more contact points than Haswell and a very different array of resistors on the bottom of the package.

The samples making the rounds are said to have Turbo clocks up to 2.9GHz and TDP ratings as high as 95W. The Google translation is a little rough, but it also mentions support for both DDR3 and DDR4 memory. Sadly, "the first batch of processors are not overclocked version." Enthusiasts may have to make do with a K-series refresh based on Broadwell.

At IDF last month, Intel said it expected to start mass-producing Skylake silicon in the second half of next year. The VR-Zone story suggests Skylake-S could arrive as early as the second quarter, but I'm not sure I'd trust unnamed sources over an official statement from the chip giant. Thanks to SH SOTN for the tip.

0 responses to “Reports: Broadwell-E slips to 2016, but Skylake-S sampling already

  1. i wonder when will we see an i5 – 6 core processor (my dream) or at least an unlocked i3 – k version –

  2. “Sure it is. Do you think Intel developed Broadwell for FREE!? Of course not!”

    Well duh. But it’s not like they have that much more to gain by introducing something based on a slightly older piece of tech when something new on the exact same process already exists.

    “How much faster do you think Skylake will be when they are offering Broadwell-K(higher cost, better performing) with Skylake-S at the same time? I doubt it’ll be much more than 10%.”

    Broadwell-K won’t have DDR4 though. Anyone who wants both will just wait the extra few months for Skylake-K.

    “That surely did not prevent Intel from giving us Ivy Bridge E, which was barely faster than Sandy Bridge E. And people get it thinking EIGHT cores, and QUAD CHANNEL memory, even though the benefit is practically zero for them.”

    Ivy-Bridge-E didn’t have any 8-core variant. It was nothing more than a speedbump outside of Xeon. And this was all for a two year (!) wait.

    “Lynnfield, and quad core mobile Clarksfield didn’t get the refresh at 32nm”

    They did. i5-760, i7-875K, 880, 740QM, 840QM and 940XM all came out in mid-2010. Look it up.

    “Arrandale/Clarkdale didn’t have a 45nm Nehalem-based predecessor”

    I know. IIRC intel had dual cores planned for Nehalem, but then scrapped them in favor of getting started with Westmere. It’s irrelevant to what I’m saying here.

    “Core i3/i5/i7 branding, meaningless processor numbers, artificial segmenting and crippling features like VT. What do you they are for? To make it easier for you? No, to lead people in the direction they want to make people buy and get more MONEY. Heck, some people think the reason they went to a crappier thermal solution in Haswell is to save a few cents! Even that may be paranoia its very plausible when a company like Intel sells nearly a million chips per day!”

    Again, I’m aware that they try to monetize in whatever way possible. Here, they’re just needlessly perseverating.

  3. [quote<] They tell you the gains are double with things like Tri-Gate than "usual" but 10 years ago those gains were had without being at useless voltages and without having Tri-Gate and such. So rather than having the product sell itself, the marketing guys step up and say we have "14nm", "Tri-Gate", "mobile-focused" and "Core i7". [/quote<] Ten years ago we had completely different scales and didn't have to worry that much about tiny issues like quantum tunneling and other fun. Also back then objectives were completely different. And BTW Trigate was being developed by Intel for ten years. Hm, wonder why they went there. And why others too including IBM, TSMC and co. (IIRC at least three different types of 3D structures) You are comparing incomparable. Complaining about wrong stuff and misattributing things. ETA: Fixed bloody tags.

  4. Sure it is. Do you think Intel developed Broadwell for FREE!? Of course not!

    ” and more cores (Broadwell-E), there’s little reason for the latter,”

    Not to mention that Broadwell-E has same number of max cores as Haswell-E at 8, because E chips are based on server EP chips they are delayed. They save MONEY by making it from very profitable server chips.

    “between new architecture”

    How much faster do you think Skylake will be when they are offering Broadwell-K(higher cost, better performing) with Skylake-S at the same time? I doubt it’ll be much more than 10%.

    “especially given that it won’t be too much of an improvement”

    That surely did not prevent Intel from giving us Ivy Bridge E, which was barely faster than Sandy Bridge E. And people get it thinking EIGHT cores, and QUAD CHANNEL memory, even though the benefit is practically zero for them.

    “And desktop quad cores were skipped completely for Westmere. (Lynnfield instead received a refresh.)”

    Lynnfield, and quad core mobile Clarksfield didn’t get the refresh at 32nm
    Arrandale/Clarkdale didn’t have a 45nm Nehalem-based predecessor

    “Not sure if it’s just money.”

    Core i3/i5/i7 branding, meaningless processor numbers, artificial segmenting and crippling features like VT. What do you they are for? To make it easier for you? No, to lead people in the direction they want to make people buy and get more MONEY. Heck, some people think the reason they went to a crappier thermal solution in Haswell is to save a few cents! Even that may be paranoia its very plausible when a company like Intel sells nearly a million chips per day!

  5. Not sure if it’s just money. Broadwell and Skylake are on the same die process. And desktop quad cores were skipped completely for Westmere. (Lynnfield instead received a refresh.)

    Plus, when it comes down to a choice between new architecture (Skylake-S/K) and more cores (Broadwell-E), there’s little reason for the latter, especially given that it won’t be too much of an improvement over the already released predecessor (Haswell-E) in all likelihood.

  6. jdaven,

    actually I thought of this as a positive thing, if this is any way near to final clock speeds.

    They can do a mini “Core uArch” transition where they lower clock speeds significantly again(4GHz to 3GHz) and tweak circuits to meet that and improve performance/clock so perf/watt improves more than that. Go from 16 stages to 12 stages and perf/watt improves 40%. 5-10% improvement with 33% less clocks.

    It would do two things much better:
    -Address ARM guys especially like Apple where they have chips at super low clocks having enormous IPC
    -Fit new process tech paradigm where the “big gains” are at low voltages, thus frequencies

  7. Some people believe tech companies are there for altruistic reasons, or to just “advance humanity”.

    No, there are there purely to make money. Much as possible. Skipping Broadwell completely indicates their revenues/investment $ is cut essentially by 50%. That makes zero sense when often companies like these care about mere 1% earnings difference. They are bean counters for a reason you know.

    If you look at Intel’s Investor Meeting PDFs, you will notice that the reason they have ~2 year cycle for process technology all has to do with “optimizing” that revenues/investment $s. If they do it too fast, it doesn’t “squeeze” enough dollars out of a product and spends way too much money to get it out faster, and if you are too slow people get less interested and buy less.

    Back in 2002, when 0.13u process Pentium 4’s were out power use went down drastically and performance went up a decent amount. I mean, it went from 1.7GHz at 80W to 2.0GHz 54W. Nowadays, BILLIONS of dollars spent on Tri-Gate and 22nm and 14nm is claimed to get 35-40% increase without telling you the little detail that it only works in very low threshold like voltages at 0.7v or something. They tell you the gains are double with things like Tri-Gate than “usual” but 10 years ago those gains were had without being at useless voltages and without having Tri-Gate and such.

    So rather than having the product sell itself, the marketing guys step up and say we have “14nm”, “Tri-Gate”, “mobile-focused” and “Core i7”.

  8. Broadwell-E’s going to be just as relevant, because extra cores on the server side more than compensates for any difference in perf/clock for server/workstation chips, and E’s are just a variant of those workstation EP chips.

    Desktop on the other hand had no changes since Haswell in 2H 2013. They got Haswell “Refresh” instead, so they can skip Broadwell(because Broadwell was so far back just few months differentiates from Skylake).

    The thing is while for decades process and architectural changes have been so relevant that next generation basically obliterated the previous generations, nowadays(despite companies like Intel telling you otherwise), things like “next generation architecture” and “new process” doesn’t do enough to be more important than things like satisfying schedules and earnings.

    I’m telling you now that whether its “Tick” or a “Tock” the improvements will be so small that they can use Tick AND Tock in one generation and even if Intel artificially segments them you wouldn’t really notice them. Which is what Broadwell and Skylake will turn out to be. Non-K versions of Skylake, no matter how much it wishes won’t ever beat Broadwell-K, because Intel isn’t a stupid company(and companies are there solely to make money).

  9. I like the mobile focus. Something I always wanted years ago: for new chips to both improve performance and decrease heat/noise and power consumption, instead of just maxing performance, usually along with increased heat and power consumption.

  10. Broadwell E slipping is not great since that part of their business is so premium but generally so far behind what their consumer offerings have in many regards, it is frustrating.

    Skylake… I’m slightly confused by what is going on there.

    I’m curious if and when we will see a consumer level 8 core from intel. (like a 6770k with hyper threading in a few years or something)

    AMD has been absent from the PC for years now so Intel only competes with its own market saturation at this point.

  11. That’s not true. At 315 mm^2, each full-sized Piledriver die is only a little bit smaller than the 356 mm^2 die for a full-blown Haswell-E.

    So in a sense, you could say that AMD has amazing expertise at selling large dies for low-low prices. Not sure about how profitable that is though.

  12. These two storied pretty much follow Intel’s market focus: mobile. Skylake has the expectation of both raising IPC and lowering power consumption further. I’m not surprised that we haven’t heard about SkyLake-Y either as that is expected to move to a SoC style design with fully integrated on-die IO (ie no chipset).

    Skylake-E on the other hand doesn’t have much competition. Broadwell-E is mainly going to be a core count and clock speed bump over Haswell-E. Without AMD being competitive in this space, Intel’s biggest competitor is what they shipped the last few generations. Wise to spread out releases to so that inventories in the market can be controlled.

    The interesting bit is the note about locked processors. Typically engineering samples that’d be passed around during this time frame would have their multipliers unlocked so that it can reasonably emulate a range of chips which Intel is considering for release. I wonder if this has changed.

  13. [quote<] Sadly, "the first batch of processors are not overclocked version." Enthusiasts may have to make do with a K-series refresh based on Broadwell. [/quote<] So... Artificially delay the unlocked Skylake processors just to keep Broadwell relevant? Yup, that sounds about right.

  14. In AMD’s defense (not jdaven’s), the early Bulldozer ES parts were clocked much lower than the final products that came out the door. Now, Bulldozer had plenty of other problems, but raw clockspeed wasn’t one of them.

  15. I’m going to downthumb you openly, just for getting upset about the thumb rankings! (you’re welcome)

  16. [quote<]Those sources provide a picture of one of the LGA1151 chips, which has more contact points than Haswell and a very different array of resistors on the bottom of the package.[/quote<] Those are capacitors.

  17. So, the chips are from the first engineering sample stepping – and you expect these to have production-level clockspeeds because why?

  18. Fascinating how jdaven’s reality distortion field has magically made it so that early engineering samples of parts are now the be-all end-all of clockspeeds.

    Considering that Haswell engineering samples were clocked [b<]lower[/b<] than this Skylake part* you obviously are showing even more ignorance about how microprocessors are actually developed in the real world. As for a 95 watt TDP... care to show any proof of those chips actually burning that much power? Yeah, didn't think so. TDP is a label, not a measurement of an actual chip running actual tasks. As we've already seen from Broadwell, Intel's 14nm process brings some pretty frightening performance levels into low power envelopes. Oh and yes, that was me who down thumbed you, because you deserved it. * proof: [url<]http://www.tweaktown.com/news/28234/first-benchmark-results-on-intel-s-upcoming-haswell-cpu-released-by-oclab-ru/index.html[/url<] 2.8GHz for a Haswell engineering sample? OBVIOUSLY NO HASWELL CHIP WILL EVER BREAK THE 3GHZ BARRIER! EINSTEIN PROVED IT!!

  19. Intel’s first 22nm processors, Ivy Bridge, came out in April of 2012 (TR reviewed the 3770K on April 23, 2012).

    Ivy Bridge-E came out in September of 2013 (TR reviewed on September 3, 2013)

    That’s roughly 16 months.

    So given that we are seeing the first consumer Broadwell parts coming out this month (October 2014) and that Intel has clearly changed priorities to push mobile Broadwell first followed by other segments, an early to mid 2016 launch for Broadwell-E is not really too out of kilter.

    [EDIT: HEY I WONDER WHO IS ANONYMOUSLY DOWTHUMBING ME!! OH GEE OH GOSH!
    I’M SURE [url=https://techreport.com/news/27246/amd-cuts-a-series-desktop-processor-prices?post=858296<]THESE GUYS[/url<] would never do that!]