Rumor: Geekbench entry hints at 16-core Core i9-7960X performance

Competition between AMD's hotly-anticipated Ryzen Threadripper and the high-thread-count chips in Intel's Core X family is sure to be the hot topic of late summer 2017. Intel is playing its cards pretty close to the vest and has yet to disclose the clock speeds and cache structures of most of its Core i9 lineup. A recent entry in the Geekbench database may provide some insight into that information as well as an early (and very rough) comparison to AMD's 16-core, 32-thread silicon. Track down your wintertime supply of calcium chloride, because it's rumor time.

The database entry in question is for a Skylake-E/EP chip running on an Asus Prime X299 Deluxe motherboard. The CPU has 16 cores and 32 threads. Each core has the expected 32 KB for holding L1 instructions, 32 KB of L1 data cache, and 1 MB of L2 cache. Intel hasn't revealed the clock speed or amount of L3 cache for the Core i9-7960X, but the CPU in the Geekbench entry runs at a base clock 2.5 GHz and has 22 MB of L3 cache. For comparison's sake, the 12-core, 24-thread Core i9-7920X has a 2.9 GHz base clock and 16.5 MB of L3 cache.

While it might be tempting to look across the field and see how early samples of AMD's Ryzen Threadripper CPUs are doing in the Geekbench database, numbers for those chips are so far not exceeding the performance of a Ryzen 7 1800X—clearly a sign that the engineering samples of those chips that have ended up in the database aren't ready for prime time yet. We can, however, establish a rough comparison between the posted Core i9-7960X scores and existing figures for the 10-core i9-7900X. The single sample of the Core i9-7960X earns a single-threaded score of 5238, roughly 4% lower than the 5446 median score for the Core i9-7900X. As for multi-threaded comparisons, the Core i9 posted a score of 33,672, almost the same exact figure as the 33,699 median for its i9-7900X brother. This latter result suggests that Geekbench may not be utilizing the 7960X's 16 cores to their fullest potential.

Intel has already released its X299 motherboard chipset and several of Core i9-7960X's smaller siblings. AMDs's Ryzen Threadripper is coming some time next month. August is certainly shaping up to be an interesting time for well-heeled PC enthusiasts and workstation builders.

Comments closed
    • ronch
    • 3 years ago

    I expect this to retail well beyond $1K. Maybe even $2K. So even if Threadripper doesn’t quite match up to it in terms of single thread but goes for the rumored $1K, anyone who thinks the $1K saved by going with AMD is still a nice sum of money will probably end up with Ryzen. Then again if you’re out for these kinds of processors you probably run software with tons of threads (otherwise why even go for so many cores?), and Ryzen has demonstrated that it somehow manages to scale a bit better when you multi-thread.

      • kalelovil
      • 3 years ago

      Prices have already been announced. Second link in this article is:
      [url<]https://techreport.com/review/31986/intel-core-x-series-cpus-and-x299-platform-revealed[/url<] (US$1700 tray price for this 16 core)

    • Krogoth
    • 3 years ago

    Thermal limitations are probably the issue. The ES probably ran on air cooling and had to keep clockspeed relatively low when compared to its lesser Skylake-X siblings.

    Not everything scales to 100% when you throw in more cores either.

    • Sahrin
    • 3 years ago

    [quote<]Threadripper CPUs are doing in the Geekbench database, numbers for those chips are so far not exceeding the performance of a Ryzen 7 1800X—clearly a sign that the engineering samples of those chips that have ended up in the database aren't ready for prime time yet. [/quote<] OK. [quote<]As for multi-threaded comparisons, the Core i9 posted a score of 33,672, almost the same exact figure as the 33,699 median for its i9-7900X brother. This latter result suggests that Geekbench may not be utilizing the 7960X's 16 cores to their fullest potential.[/quote<] Huh? I think you meant to say, "the 7960X's 16 cores are not ready for prime time yet." That's the conclusion we're drawing about unreleased products, right?

      • Waco
      • 3 years ago

      This. The wording between the two has a subtle bias that I don’t expect to see in TR reports.

        • Jigar
        • 3 years ago

        I second that.

          • Shobai
          • 3 years ago

          I can only agree, Jeff.

          • the
          • 3 years ago

          I concur as well.

          If Geekbench behaves similarly with poor thread scaling across both platforms, I would suspect that Geekbench is the one with the issue.

      • psuedonymous
      • 3 years ago

      [quote<]Huh? I think you meant to say, "the 7960X's 16 cores are not ready for prime time yet." That's the conclusion we're drawing about unreleased products, right?[/quote<] Given the single-threaded score is within the margin of error of the i9-7900X, assuming a core utilisation issue is a reasonable assumption when the multi-core score is also nearly the same.

        • synthtel2
        • 3 years ago

        But if so, that’s a good hint that the same thing might be happening with Threadripper.

    • chuckula
    • 3 years ago

    Only related in a general sense, but Cinebench has been “broken”

    [url<]https://www.servethehome.com/cinebench-r15-is-now-a-broken-as-a-benchmark-and-11-5k-surpassed/[/url<]

      • Krogoth
      • 3 years ago

      It is more like it was never coded to handle that many cores in the first place. I also suspect that cache coherence and NUMA-related issues are creeping up here (common problem on 4P platforms). Besides, it is kinda silly to run a workstation-tier benchmark on a HPC/high-end datacenter server platform.

    • chuckula
    • 3 years ago

    If you look at the full listing you’ll see results that are all over the map due in part to overclocking but also just to abject randomness in how Geekbench runs.

    For example, if you sort by multi-threaded scores you’ll see the fastest 7900X gets over 43000 in multithreaded. A YUGE overclock you say? Maybe, but look at the single-threaded score for the exact same chip: it’s only 5380 that is actually [i<]lower[/i<] than the "average" score TR posted above... that's some bizarre overclock. In other words: I look at Threadrippers low Geekbench scores as a positive because Geekbench is a lousy benchmark.

      • derFunkenstein
      • 3 years ago

      Geekbench is just not worth talking about, for sure. It would have you believe that an iPhone 6S is about 50% as fast as any of these Skylake-E CPUs in single-threaded loads. All you have to do is load up Safari, which is [i<]heavily optimized[/i<] on the iPhone and Chrome on a desktop to see it's not really even that close.

        • tipoo
        • 3 years ago

        Or that the A10X is 1/3rd of i9s performance at 1/28th, or 3.57%, of its power draw.

        I do believe Apple chips are crazy fast for their space, but with that above line and yours, something’s fishy, on Geekbench’s side.

        If these were completely face value, well, Apple would just slaughter the server space, then the laptop and desktop space after, selling chips alone, lol.

          • derFunkenstein
          • 3 years ago

          Apple is delaying my plans for a Geekbench cluster by not moving Macs to ARM.

          • NoOne ButMe
          • 3 years ago

          Scaling clocks is hard.
          Also where is powerdraw numbers from 🙂

          Have not seen a total breakdown of power from Apple’s SoC long time. Love go see one!

            • tipoo
            • 3 years ago

            Coming from my assumptions, you know what they say about that 😛

            But I don’t expect it’s past a 5W chip, maybe 10W at peak? It’s in a form factor that Core M would get challengingly warm in, and it’s only lukewarm at full tilt.

            • NoOne ButMe
            • 3 years ago

            IP7+ heats up, but maybe that is wifi/cellular happening…
            Maybe GPU, remember initial reviews saying GPU got hot… but why hot when just streaming music (25-30C ambient)

            I guess 1.5-2.5W per core depending on how loaded is at 2.25Ghz.

            Like how Intel and AMD cores power fluctuate depending on load.

        • guardianl
        • 3 years ago

        To be frank, I’m not sure the subjective test of two wildly different browsers on two different OSes means anything.

        [url=http://www.anandtech.com/show/9766/the-apple-ipad-pro-review/4<]SPEC2006 scores for A9X etc.[/url<] TLDR; The Perl/GCC sub scores* indicate the A9 CPU has approximately broadwell-level IPC. So yes, in some single thread workloads, the A9 really is 50% as fast as Skylake-E with the performance difference mostly in clock speed (~2Ghz vs~4Ghz). * SPEC isn't perfect either, but the GCC, Perl and XML tests are pretty decent representations of workstation workloads.

          • derFunkenstein
          • 3 years ago

          The two different browsers on two different OSes at least represents a real-world task. Maybe after I build my Geekbench cluster I’ll work on a SPEC benchmarking farm.

Pin It on Pinterest

Share This