Intel reveals full Core i9 specs and launch dates

Intel took the wraps off the remainder of its Core i9 CPU lineup today. We now have full specs for the Core i9-7920X, i9-7940X, i9-7960X, and the i9-7980XE. Here's the Core i9 lineup in its place with the rest of Intel's Core X CPU family:

Model Base

clock

(GHz)

Turbo

clock

(GHz)

Turbo

Boost

Max 3.0

clock

(GHz)

Cores/

threads

L3

cache

PCIe

3.0

lanes

Memory

support

TDP Socket

Price

(1K

units)

i9-7980XE 2.6 4.2 4.4 18/36 24.75 MB 44 Quad-channel

DDR4-2666

165W LGA 2066 $1999
i9-7960X 2.8 16/32 22 MB $1699
i9-7940X 3.1 4.3 14/28 19.25 MB $1399
i9-7920X 2.9 12/24 16.5 MB $1199
i9-7900X 3.3 4.5 10/20 13.75MB 140W $999
i7-7820X 3.6 8/16 11MB 28 $599
i7-7800X 3.5 4.0 NA 6/12 8.25MB Quad-channel

DDR4-2400

$389
i7-7740X 4.3 4.5 NA 4/8 8MB 16 Dual-channel

DDR4-2666

112W $339
i5-7640X 4.0 4.2 NA 4/4 6MB $242

These specs are identical to those in a recent leak. I'd love to offer some deep commentary, but the story really seems quite simple. Performance across all of these parts should be impressive. Some might be concerned by the rather low base clocks on some of the higher-core-count parts, but those figures likely reflect the most pessimistic scenario for AVX-512 workloads. My real-world experience with Skylake-X suggests Intel's Turbo speeds are plenty attainable under non-AVX workloads, and the Turbo speeds for the 7920X, 7940X, 7960X, and 7980XE are just as aggressive as those of the lower-TDP i9-7900X.

Intel seems to be launching these chips on a more aggressive pace than industry chatter seemed to suggest. The Core i9-7920X will be available August 28, and the Core i9-7940X, Core i9-7960X, and Core i9-7980XE will be available September 25. Pricing for those higher-core-count CPUs hasn't changed in the shadow of the Ryzen Threadripper launch, either, suggesting Intel remains confident in the performance-per-dollar of its platform.

In any case, the battle for high-end desktop supremacy seems set to heat up as the temperatures begin to fall. We can't wait.

Comments closed
    • NoOne ButMe
    • 2 years ago

    Anandtech got the turbo boost numbers from PCGamer into a chart, and it makes things more confused….
    [url<]http://images.anandtech.com/doci/11698/turbos.png[/url<] I'm guessing most of the turbo numbers for high number of active cores will only be usable on exotic cooling maybe? I suspect the TPD is for base clocks only, and turbo on over a few cores will only work if you have huge overkill near-exotic water cooling? [on an aside, 10C/14C still look like the best two SKUs if you need over 28 PCIe lanes and prefer Intel]

      • Mr Bill
      • 2 years ago

      I posted a link to those below and I think that table is very expressive. It shows very nicely how frequency has to be laid off to lower numbers as the number of cores goes up. It also shows how for the same number of cores, a faster bin on the X-axis can have higher frequencies at each step.

    • bjm
    • 2 years ago

    Has there been any new updates regarding Turbo Boost Max 3.0 and Linux support?

    • floodo1
    • 2 years ago

    7820X and 7800X seem interesting, what’s the word?

      • ImSpartacus
      • 2 years ago

      They are bad for gaming (increased intra-die latency due to new mesh), but otherwise high performing.

      However, for the money, if you don’t care to game, Threadripper and the single socket Epyc parts look pretty potent.

        • Klimax
        • 2 years ago

        If most of threads are independent and no cross-thread communication. Otherwise there will be trouble. (Without NUMA-aware code) But even in that optimal case one has to be careful about that lack of IPC in Ryzen.

    • TwoEars
    • 2 years ago

    Still waiting for that 5 GHz stock speed.

    • mganai
    • 2 years ago

    Little typo on the table: only 7940X and up are 165W TDP. 7920X is 140W.

    • PrincipalSkinner
    • 2 years ago

    Coffee Lake just called. Wants to know what’s up with 7800X and below.

      • chuckula
      • 2 years ago

      That’s akin to saying that the R7 1700 – 1800X called and wants to know why there’s a more expensive Threadripper 1900.

        • sreams
        • 2 years ago

        Threadripper 1900 offers all of the PCI Express lanes and 4-channel memory that the larger CPUs do. That’s why.

          • chuckula
          • 2 years ago

          The 7800X offers quad-channel memory support and 28 lanes of PCIe at a substantially lower price point than many of AMD’s AM4 products (much less threadripper) too.

          As for the OMG 60 lanes of PCIe* that’s a wonderful fanboy talking point but since literally nobody screamed about the 1800X only having 16 lanes of PCIe was some major drawback, it’s just that: A talking point that feels good but has little or no real-world implications.

          * If you say 64 you’re wrong. It’s 60 because 4 are taken up by the chipset.

            • NoOne ButMe
            • 2 years ago

            Than many? You mean all AM4 (to date) CPUs! I never got why AMD limited it, as clearly Ryzen has 32 lanes.

            And given the 8C TR exists, well. $150, maybe less depending on Mobo prices (I imagine about equal for X299 and X399) for an extra 32/36 PCIe lanes (does x299 take 4 for chipset?).

            More importantly, it has enough lanes to go fully x16 and have plenty left over for expansion. Where 28 is a tad to low for some. If they need GPU on a true x16 is a whole ‘nother story.

            If Intel had stayed at 44 the whole line, this would be a non-issue. I hope Intel loses more in sales for low PCIe lanes than they gain from upwelling people.

    • NoOne ButMe
    • 2 years ago

    so the Intel lineup is basically:
    7800X
    7900X
    7940X
    7980XE

    I don’t see a part beyond the first three which warrants a purchase/consideration. The fourth only selling for the people who go “oh, it’s the highest price/most cores, must be the best”.

    The 7800X because it brings higher ST at stock, and MT should be mixed bag or outright win at stock/overclocked.

    7900X, because it brings the PCIe lanes, and is the highest (stock) clocked part with 44 lanes.

    7940X because compared to the 7900x, it only loses a tiny bit of clockspeed for the 4 extra cores.

      • mcarson09
      • 2 years ago

      7980XE will probably come in Xeon form. With the 128GB memory cap on both Manf. HEDT platforms, it’s better to look at the server parts coming up.

    • leor
    • 2 years ago

    What do they mean by available? Is the 7900x supposed to be available now? Cause it isn’t.

      • chuckula
      • 2 years ago

      [url<]https://www.newegg.com/Product/Product.aspx?Item=N82E16819117795&cm_re=i9_7900x-_-19-117-795-_-Product[/url<] [url<]https://www.amazon.com/Intel-BX80673I97900X-Core-i9-7900X-Processor/dp/B072KTSCCS/ref=sr_1_2?s=pc&ie=UTF8&qid=1502135849&sr=1-2&keywords=i9+7900X[/url<]

        • leor
        • 2 years ago

        Those just showed up, and still above list price. They have been out of stock since the “launch date,” and where you could find one, they were being sold by third parties for $1,500.

        Not sure why I was down voted, I’m looking to buy one of these, and have been checking the stock every few days. It’s good to see them starting to trickle into stores, but at this point I’m going to wait for the Thread Ripper reviews before I pull the trigger.

          • chuckula
          • 2 years ago

          They have been in somewhat tight supply and demand has been strong.

          At least the price premiums aren’t at Rx-580 levels.

          • kuraegomon
          • 2 years ago

          Apparently things are slightly better up here in Canada:

          [url<]http://www.canadacomputers.com/product_info.php?cPath=4_1210_65&item_id=108917[/url<] I speak only of stock levels. If you factor in exchange rates, that Newegg.com listing is slightly cheaper. I'm surprised the delta isn't worse actually.

    • HTarlek
    • 2 years ago

    With 4 cores running at 4.3 GHz, that i7-7740X should be an excellent gaming platform.

      • RAGEPRO
      • 2 years ago

      Rather pay less and get an i7-7700K, which is literally the same CPU on a more sensible platform.

        • chuckula
        • 2 years ago

        Screw either of those options.

        Covfefe lake or GTFO.

    • Wildchild
    • 2 years ago

    [quote<] 24.75 MB of L3 cache [/quote<] Holy cow.

      • ClickClick5
      • 2 years ago

      [url<]https://en.wikipedia.org/wiki/POWER7[/url<] In 2010, their 8 core model racked 32MB of L3. That blew me away when nit came out. And then POWER8 in 2013, and now POWER9, with 120MB L3.

        • UberGerbil
        • 2 years ago

        The [i<]dual[/i<]-core [url=http://ark.intel.com/products/28003/Intel-Itanium-Processor-9050-24M-Cache-1_60-GHz-533-MHz-FSB<]Itanium 9050[/url<] (Montecito core) carried 24MB of L3 cache and almost [url=http://www.realworldtech.com/super-heavyweights/4/<]27MB total[/url<] (L2+L3) cache... in [url=http://www.realworldtech.com/montecito-launch/<]2006[/url<]. That was mind-blowing at the time. Of course it needed it to put up competitive numbers.

        • the
        • 2 years ago

        IBM likes to use eDRAM for last level cache on-die to achieve such capacity. IBM still uses SRAM for the L1 and L2 caches which are more latency sensitive in the cache hierarchy.

        Intel still favors SRAM, Crystalwell being an exception. It is interesting that Intel is continuing this trend with Skylake-X/SP/EP/EX/OMGWTFBBQ. The L3 cache is barely bigger than the L2 cache now but due to exclusivity, this isn’t too much of a burden. I am curious what things would have been like if Intel used an eDRAM cache for the L3, increased L3 capacity by a factor of 4 and kept the caches inclusive for coherency simplicity.

      • Tristan
      • 2 years ago

      Just 1.3 MB / core, pretty small. 7700K have 2mb / core

        • the
        • 2 years ago

        This is offset by the larger 1 MB of L2 cache per core which is exclusive of the L3 cache. So basically it is 2.375 MB of combined L2 + L3 cache per core.

        While the i7 7700K has 256 KB of L2 cache per core, all of its contents exist in the L3 cache due to its inclusive nature.

    • ludi
    • 2 years ago

    Looking forward to an eventual i9-7900X or 7920X comparo against the TR-1950X. How much performance do you actually get at $1000?

      • kuraegomon
      • 2 years ago

      This. i9-7900X vs TR-1950X is exactly what I’m waiting on before I replace my 3930K build.

        • Klimax
        • 2 years ago

        If SSEx at most and avoiding cross-thread communication, it’ll be +- similar. Anything else will go likely to Skylake-X.

      • Hattig
      • 2 years ago

      Yeah, actually let’s factor in the motherboard cost as well.

      They both quad-channel RAM IIRC, so that’s equal.

        • psuedonymous
        • 2 years ago

        Size is also a factor: there is at least one Skylake-X ITX board, but as far as I am aware nobody has even announced an ITX Threadripper board (and with the size of the socket, it may not even be possible).

    • chuckula
    • 2 years ago

    Meh!

    Unless Intel can come up with insanely over the top product packaging and then ship out review samples in the same type of suitcase preferred by 9/10 assassins with individual numbering and the names of Youtube blogs laser-etched onto completely non-functional heatspreaders then they clearly are trying to use their evil monopoly powers to skew the reviews.

    Oh, and unless Intel intentionally snubs TR and refuses to ship them a review unit followed by a public backlash then I’m going to say that Intel marketing clearly has it in for TR.

      • sreams
      • 2 years ago

      “Intel marketing clearly has it in for TR”

      Well, of course it does. Threadripper is direct competition for these Core i9 CPUs.

      😉

    • ImSpartacus
    • 2 years ago

    [quote<]Performance across all of these parts should be impressive. Some might be concerned by the rather low base clocks on some of the higher-core-count parts, but those figures likely reflect the most pessimistic scenario for AVX-512 workloads. [/quote<] I don't think Intel advertises AVX clocks (be that for AVX-512 or otherwise) in typical marketing material. The AVX clocks are usually really low due to the power consumption of the big vector units (though actual performance is still high since these vector units are beefy af). Anandtech has a sick graph of the various turbo clocks for a particular Xeon chip in the various AVX modes. As you'd expect, AVX clocks are very low. [url<]http://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade/8[/url<] Honestly, these turbo curves get way less attention than they should. Now that Intel is moving to 6C in mainstream via Coffee Lake, we need to pay more attention to the actual turbo curve. This trend of only caring about the base clock and 1C turbo is getting to be unreliable since a typical consumer workload rarely stresses only one core and a typical consumer workload rarely includes a power virus-esque load to utilize the base clock.

      • smilingcrow
      • 2 years ago

      The Wikipedia pages usually show all the Turbo speeds for all the various core variations at least for Xeons so you can quickly get a sense of it.
      It’s mainly common sense so a combination of core counts and TDP gives a good idea of what to expect.

        • ImSpartacus
        • 2 years ago

        Where exactly are they on Wikipedia?

        I see turbos for the consumer stuff. That’s easy, just 1C, 2C, 4C, and base.

        And then on the Xeon stuff, I see a “normal” and “turbo” for non-avx loads. It’s not that simple.

        The problem is when you’ve got a CPU with 10+ cores, you have several dozen unique turbo clocks. It’s not going to fit into the classic table.

        Seriously, check out the link I shared. There are A LOT of turbo clocks for the big Xeons.

          • smilingcrow
          • 2 years ago

          Yes, they list it in one string like this for example 1,1,2,2,3,3,4,5,5,6,7,7 so giving the max turbo boost for every combination of cores.
          I haven’t looked for the current generation but always found it for previous generations. There may be a delay until someone digs through the documentation to get the info.

            • ImSpartacus
            • 2 years ago

            I’m not seeing it in Skylake, Broadwell or Haswell and I don’t recall ever seeing it on earlier articles.

            [url<]https://en.m.wikipedia.org/wiki/Skylake_(microarchitecture)[/url<] [url<]https://en.m.wikipedia.org/wiki/Broadwell_(microarchitecture)[/url<] [url<]https://en.m.wikipedia.org/wiki/Haswell_(microarchitecture)[/url<]

            • smilingcrow
            • 2 years ago

            Try this:
            [url<]https://en.wikipedia.org/wiki/List_of_Intel_Xeon_microprocessors#.22Broadwell-EP.22_.2814_nm.29[/url<] They don't have the data for all the Xeons and nothing for Skylake yet.

      • chuckula
      • 2 years ago

      Looks OK to me.

      [url<]http://x265.org/x265-receives-significant-boost-intel-xeon-scalable-processor-family/[/url<]

        • DavidC1
        • 2 years ago

        This is the very optimistic scenario with AVX-512 optimizations though.

        It seems reactions on SKL-X is quite mixed. Because enthusiasts need both high single thread performance and want high amount of cores.

        If you want a “true” enthusiast core you’d hope Intel to scale up the core counts for the regular LGA115x socket without using the mesh.

          • chuckula
          • 2 years ago

          Really, it seems like a pretty realistic test case with an incredibly popular video encoder.

          Those numbers aren’t showing 100% increases in performance and it’s not an artificial synthetic workload.

          • the
          • 2 years ago

          The previous generations of Xeon’s would be an indicator of how high Intel can scale the mesh: 14 stops on the ring bus is the highest they’ve used. The hard limit for the ring bus I would [i<]guess[/i<] to be 16*. Note that items like PCIe controller, QPI link, memory controller and GPU all take a stop on the ring bus too. So for a socket 115x product, Intel could go all the way up to 12 core and be under that 16 hypothetical limit with out adding a second on-die ring bus. *It isn't clear if the 16 core Xeon-D's have one or two ring buses. If a single ring, then the technical limit is far higher on a ring bus but latency probably necessitates splitting the bus up.

        • ImSpartacus
        • 2 years ago

        I don’t understand. What’s ok?

          • chuckula
          • 2 years ago

          A 30 to 60% performance boost from Broadwell in a real-world encoding application that’s regularly used in industry for real-world workloads.

          Looks like the low clockspeeds aren’t that much of an issue.

          Oh and that’s on a Xeon part that doesn’t turbo the AVX-512 anywhere nearly as aggressively as overclocked HEDT chips do. A properly overclocked 7900X can usually run all 10 cores in AVX-512 at around 4GHz or so.

            • ImSpartacus
            • 2 years ago

            I never said low clock speeds were an issue.

            Even on reduced AVX clocks, the amount of work being done is still significant due to the sheer width of the vector units.

        • the
        • 2 years ago

        I’m not sure how much of that is just pure AVX-512. The gains discussed were in the context of a per core basis but there is a ~14% difference in clocks which eats up a good chunk of the gains. The Sky Lake-EP chips also have the benefit of both faster memory and more memory channels. There is also a slight uptick in performance per clock from Broadwell to Sky Lake outside of AVX-512. The raw benefit of AVX-512 is there but it doesn’t seem to be that great when you consider other improvements.

        The good news is that per chip, Intel has increased core counts once again for in aggregate there is a very respectable performance gain.

      • xeridea
      • 2 years ago

      Ya I would say 7940X would be best chip to get. Reasonable base, and 14 cores for $1400. At base clocks, 7980 XE is only 8% faster, for $600 more.

      • blastdoor
      • 2 years ago

      Another dimension is time.

      Say you’ve got the 7960X running all cores for half an hour, non-AVX load.

      Is there any realistic chance that you’ll be seeing a clock speed above the base without exotic cooling?

        • ImSpartacus
        • 2 years ago

        You’re probably right.

        You don’t spend this much money on a big beefy CPU if you’re only loading half of the cores.

        Or maybe you do. I honestly don’t know much about that side of things.

        • Klimax
        • 2 years ago

        Depends on thermal transfer from chip to lid. If sufficiently good then some beefy air coolers will manage it.

      • Mr Bill
      • 2 years ago

      I’m impressed by the presentation of this table (update on 8/8) at Anandtech that he reformatted from PCGamer’t table… [url=http://www.anandtech.com/show/11698/intel-finalizes-skylakex-processor-specifications-18cores-44-ghz-165w-on-september-25th<]Non-AVX Turbo Frequencies[/url<]. It would really lay out the value proposition if the prices were included.

    • srg86
    • 2 years ago

    If I was in the market for an HEDT processor, which I’m not, Coffee Lake would be more my thing, I think i7-7820X would be the sweet spot.

    Kinda shame about the number of PCIe lanes, but otherwise looking good. No way would I spend $2000 on a CPU.

      • sreams
      • 2 years ago

      “Coffee Lake would be more my thing”

      Do we really know enough about Coffee Lake yet to decide that?

        • srg86
        • 2 years ago

        If Coffee Lake is 6 cores and integrated graphics, then yeah it would be more my thing.

        • ImSpartacus
        • 2 years ago

        We’ve had a ton of leaks, starting about a year or two ago and they’ve been very consistent.

        You can think of Coffee Lake as basically 6C Kaby Lake. It’s expected that it’ll have roughly the same clocks due to the 14++nm process improvement (except base clock obviously drops).

        [url<]https://videocardz.com/71660/intel-coffee-lake-architecture-brings-6-core-at-95w-tdp[/url<] [url<]https://videocardz.com/71307/intel-core-i7-8700k-i7-8700-i5-8600k-i5-8400-specs-emerge[/url<] We'll probably see a partial release in the next few weeks with a full release in January.

        • Klimax
        • 2 years ago

        Assuming reports of frequency of 6 core chips are correct, then yes, we do know. And it’ll be murder on all chips especially Ryzen. (Intel’s IPC * Coffee frequency = insane performance)

        Note: One reason why one might want to wait is that Skylake-X is using original 14nm per David Kanter’s posts on RWT.

    • credible
    • 2 years ago

    And what is the purpose of the bottom 2 sku’s?

      • DPete27
      • 2 years ago

      More PCIe lanes, more OC potential.
      [edit] I should add, with generally higher mobo cost than Z270.

        • smilingcrow
        • 2 years ago

        Kaby Lake X has the same number of lanes as Kaby Lake but I’m not sure how the chipsets compare for lanes?

          • juzz86
          • 2 years ago

          X299 is Z270, lanes-wise.

      • Krogoth
      • 2 years ago

      Stop-gap solutions to ensure that there’s enough Socket 2066 chips in the market for motherboard vendors to justify making Socket 2066 boards.

      It appears that Intel is having significant yielding issues with its Skylake-X chips.

      Kaby-Lake X will be abandoned as soon Coffee Lake comes around and Intel builds enough inventory of its Skylake-X chips (mainly 7800X-7920X).

        • smilingcrow
        • 2 years ago

        Just looked at 2 retailers (Scan & Ebuyer) in the UK and they both have the 6, 8 and 10 core chips in stock and Scan have most of them as OEM as well as retail so not seeing a shortage.

        Having the quad cores to justify the market is like having Celerons and Pentiums to justify the market for people who want an i7-7700K.
        The majority buying this platform care less about the quads and everyone knows that including the mobo manufacturers.

      • flip-mode
      • 2 years ago

      Stepping stones to make the platform accessible to lower budgets? Maybe? Just guessing.

    • shank15217
    • 2 years ago

    Oh look competition!

      • hungarianhc
      • 2 years ago

      Bingo!

      • willyolioleo
      • 2 years ago

      “Competition”

      > $1000 = 10 cores, 44 PCIe lanes

      > 16 cores = still only 44 PCIe lanes, $1700

      lol

        • jihadjoe
        • 2 years ago

        Yeah, it should be $1600 if they kept the price per-core the same!

        But srsly this is why I hope Threadripper makes a dent on Intel’s HEDT market share. Intel obviously won’t lower prices any further unless you given a good reason to.

Pin It on Pinterest

Share This