Intel bolsters Core X CPUs with higher clocks, more cache, and solder

Along with its refreshed Coffee Lake chips, Intel announced an update to its Skylake-X CPUs for the X299 platform yesterday morning. While the underlying architecture of these parts is the same as the first round of Core i7 and Core i9 CPUs for the platform, quite a bit has changed in the way Intel manufactures and packages those CPUs.

Base

clock

speed

(GHz)

Peak

single-core

boost

speed

(GHz)

Turbo

Boost

Max 3.0

peak

speed

(GHz)

Cores/

threads

TDP

(W)

L3

cache

(MB)

CPU

PCIe

lanes

Memory

support

Price
i9-9980XE 3.0 4.4 4.5 18/36 165 24.75 44 Four channels

DDR4-2666

$1979
i9-9960X 3.1 16/32 22 $1684
i9-9940X 3.3 14/28 19.25 $1387
i9-9920X 3.5 12/24 $1189
i9-9900X 10/20 $989
i9-9820X 3.3 4.1 4.2 16.5 $889
i7-9800X 3.8 4.4 4.5 8/16 $589

As with Intel's unlocked mainstream desktop parts, one of the loudest complaints about Skylake-X CPUs regarded the effectiveness of the paste-type thermal interface material the company began using in lieu of solder on its first run of those chips. Indeed, we've rarely found ourselves limited by the underlying silicon when overclocking our Skylake-X CPUs—the wall has been entirely thermal. With its refreshed Skylake-X CPUs, Intel has resumed soldering the die to the integrated heat spreader (or IHS). That move could finally allow overclockers to reach the limits of refreshed Skylake-X parts without the hassle of popping off the IHS and applying more effective thermal interface material of their own.

Refreshed Skylake-X parts have also moved to Intel's refined 14-nm++ process technology. The apparent benefits of this move emerge in some small spec adjustments from the first round of Skylake HEDT parts. Intel now specifies a Turbo Boost Max 3.0 speed of 4.5 GHz across the board for these chips (save for the Core i9-9820X and its victim-of-segmentation 4.2-GHz TBM 3.0 speed). Depending on the chip in question, peak Turbo Boost speeds have also increased anywhere from 100 MHz to 200 MHz (again excluding the odd-man-out i9-9820X). 

Base frequencies for these parts have also increased across the board, but those increases likely won't matter to enthusiasts installing these CPUs in well-cooled systems. Intel has also increased TDPs for all of these parts to 165 W, a move that could be an effort to allow the chips to take greater advantage of the performance on offer from the improved process. Taken together, those increases in clock speed, process, and thermal headroom could offer small but welcome improvements in performance in lightly threaded and multithreaded workloads alike. We'll have to wait until we see the full Turbo Boost tables for these parts to be sure, though—information that Intel remains stubbornly resistant to disclosing.

Another change for refreshed Core X-series chips is that L3 cache sizes on several models have increased. Skylake-X cores each have 1.375 MB of L3, and Intel can disable some of those cores on a die while leaving their shared L3 caches available to the active cores on the chip. As Ian Cutress at Anandtech points out, the fact that Intel is now enabling more cache than there are active cores on some refreshed Core X chips means the company is likely flipping switches on its high-core-count Skylake Server silicon to make these consumer parts now, rather than repackaging its low-core-count chips. That fact could also help to explain the higher TDP of refreshed Core X CPUs in general.

An X299 pain point that AMD has been probing with sharp instruments is Intel's choice to segment the number of PCIe 3.0 lanes available from the CPU. High-end desktop users who need plenty of PCIe lanes for storage or expansion cards can rejoice, as Intel is dropping its 28-lane and 44-lane product tiers. All refreshed Skylake-X CPUs can deliver 44 lanes of PCIe 3.0 connectivity directly to expansion cards and storage devices.

One X299 limitation that's sticking around is the platform's 128-GB maximum memory capacity. Intel also isn't permitting motherboard makers to add ECC RAM support to their boards, as was sometimes possible with the X99 platform. Users who need the assurance of ECC will still need to use Xeon W platforms, not Core X.

Intel will be releasing its refreshed Core X chips next month.

Comments closed
    • Ninjitsu
    • 1 year ago

    lol it’s pretty hilarious to see all these changes across the board

    • ptsant
    • 1 year ago

    Nice products, but they only make sense because the Xeons are absurdly expensive (or is it the other way around?). Best buys seem to be 9900X and 9920X

      • NoOne ButMe
      • 1 year ago

      they have much higher boost (and base) clocks compared to Xeon.

    • loonster
    • 1 year ago

    Why are these using HCC chips?
    1) Simplify logistics
    2) Intel finally has enough partially defective chips to warrant selling them as a lower bin.
    3) Intel has an excessive supply of HCC chips

      • DavidC1
      • 1 year ago

      Because HCC allows up to 18 cores.

        • jihadjoe
        • 1 year ago

        Still feels like a huge waste to be fusing off more than half the chip.

          • DavidC1
          • 1 year ago

          They are not, if they are using the same arrangement as Skylake-X.

          8-10 core uses LCC, which allows for maximum of 10 cores.
          12-18 cores uses HCC, which allows for maximum of 18 cores.

          So they are still disabling some cores, but at die sizes this big it might even be necessary.

          • NoOne ButMe
          • 1 year ago

          they are nearly 500mm^2.

          Given they have power requirements, which in datacenter will be strict, the ones we see with extra cache probably failed the power requirements for server and for consumer with higher core counts.

    • hansmuff
    • 1 year ago

    Still SKL based, so zero new mitigations against side channel attacks.

    The new desktop line at least has *some* new hardware mitigations.

    Lame.

      • Klimax
      • 1 year ago

      ETA: Misread my own link brutally. Post deleted for terminal case of wrongness.

    • jihadjoe
    • 1 year ago

    Urgh 9900X/9900K.

    Someone is eventually going to get the [url=https://i.imgur.com/2vXNfUj.png<]wrong CPU for the wrong motherboard[/url<].

      • Klimax
      • 1 year ago

      Not first by long shot and not last either. (Pentium III Tualatin for example)

      • DavidC1
      • 1 year ago

      I’ve see a picture where a guy busted his CPU AND the pins on the motherboard because he tried to force a Haswell chip to a Skylake board.

      #1 lesson in electronics. If you need to force it, you are probably doing it wrong.

    • DavidC1
    • 1 year ago

    7980XE seems to be surprisingly competitive in many reviews against the 2990WX. Many prosumer applications seems to have trouble scaling beyond certain amount of threads. There’s also the additional latency due to needing to transfer data off-die.

    If these parts can get 5-10% faster, and the solder thermal interface allows for further 10% overclocks over the 7980XE, it may not look dire for Intel as it seems. 2990WX still seems like a better deal, but not absolutely dominant.

    Neither platform is what a gamer should get of course.

      • blastdoor
      • 1 year ago

      Yes, definitely an interesting set of tradeoffs. I’m finding the memory issues with the 2990wx can really be make or break. Sometimes some careful tweaking can help circumvent the memory bottlenecks, but not always.

      I can imagine that even if Apple were to someday use an AMD CPU, it won’t be anything like the 2990wx — this is a tweaker’s CPU that is incompatible with Apple’s “it just works” design goal.

      Personally, I’d be willing to pay more for “it just works” ease of use, but Apple doesn’t compete in this space anymore.

    • wierdo
    • 1 year ago

    (moved to [url<]https://techreport.com/news/34159/core-i9-9900k-takes-eight-cores-16-threads-and-solder-to-300-series-mobos?post=1092930#1092930)[/url<]

    • Eversor
    • 1 year ago

    Typical Intel… 8 and 10 core CPUs that make no sense for HED. 12 core CPUs at substantially higher prices than what AMD will be offering soon.

    Are they just hoping that nobody notices? Core clocks certainly don’t seem to make up for it.

      • Krogoth
      • 1 year ago

      Their shareholders aren’t willing to do a price war with AMD in the HEDT market yet. Intel needs to keep up those pre-Ryzen/pre-Threadripper era profit margins.

        • Pancake
        • 1 year ago

        Shareholders don’t make those decisions.

        Why should Intel engage in a “price war”? They’re selling everything they can make and are moving their product mix higher end. It would be illogical behaviour.

          • Krogoth
          • 1 year ago

          They indirectly do make these decisions. They elect executives who have an incentive to bring as much profit as possible. If the executives fail to bring in projected figures they usually end up getting fried or demoted.

          Intel shareholders are just too used to the juicy margins for the past decade or so to suggest the executives to wage a price war in the HEDT and higher-end desktop market. They might change their tune once AMD manages to snag significant marketshare away from Intel in these segments. It will start to cut into the profit/revenue stream.

            • K-L-Waster
            • 1 year ago

            Shareholders holding execs accountable for poor execution /= shareholders dictating strategy.

      • Kretschmer
      • 1 year ago

      If they sell it all, is this a bad choice?

      • chuckula
      • 1 year ago

      That’s funny, literally nobody complains when AMD sells these: [url<]https://www.amazon.com/AMD-Threadripper-16-thread-Processor-YD190XA8AEWOF/dp/B0754JNQBP[/url<]

    • chuckula
    • 1 year ago

    AMD: We give you M0ar of the C0ar!

    Intel: We give you M0ar of everything but the C0ar.

      • drfish
      • 1 year ago

      Gah, you made me forget how to spell “core” for a few seconds today. 😛

        • chuckula
        • 1 year ago

        Chuckula: TR’s official anti-editor.

    • shank15217
    • 1 year ago

    No ECC means no X platform. You got to be kidding me, these machines hold over 64 GB of RAM and having no ECC to cover that is plain stupid. Also what is the real difference between workstation and HEDT? If you need more cores then you probably need more ram, and if you need more ram then you probably could use more pci lanes for all your peripherals.

      • srg86
      • 1 year ago

      Although nice to have, to me, no ECC = meh don’t care.

        • ptsant
        • 1 year ago

        If your workloads complete in minutes or even an hour then an occasional crash is no big deal. If you run calculations that take days/weeks you should care. The worst is when your data get corrupted. I recently had to recover data from an archive because they had been corrupted in transit.

      • cygnus1
      • 1 year ago

      Yeah, it does seem extra segment-y to not allow ECC when the underlying silicon almost assuredly supports it. I just can’t think of too many applications that would use that much memory and not want ECC to go with it.

      I mean, my desktop has 32GB but that’s just because I’m a browser tab junky and the cheapo DDR3 back in the day.

        • Waco
        • 1 year ago

        This is why Threadripper is killing it for HEDT builds.

      • Srsly_Bro
      • 1 year ago

      And what about if you need more peripherals?

Pin It on Pinterest

Share This