AMD sets date for Threadripper 2970WX and 2920X chips along with Dynamic Local Mode

Friday afternoons aren't usually fertile ground for big news. Today's a little different, though. Out of the blue, AMD told the world when the previously announced Ryzen Threadripper 2970WX and 2920X chips will be ripe for the picking. The red team's also letting the Dynamic Local Mode cat out of the bag soon. Let's rehash some basics first.

Model

Cores/threads

Boost clock (GHz) Base clock (GHz)

L3 cache (MB)

TDP

Suggested price

Ryzen Threadripper 2990WX

32/64

4.2 3.0

64

250 W

$1799

Ryzen Threadripper 2970WX

24/48

$1299

Ryzen Threadripper 2950X

16/32

4.4 3.5

32

180 W

$899

Ryzen Threadripper ​​​2920X

12/24

4.3 $649

As the table above shows, the Threadripper 2970WX could roughly be described as a 2990WX with fewer cores. The 2970WX is a 24-core, 48-thread jobbie that should hit the same clock speeds as its bigger brother. It also has the same 64 MB of cache and 250-W TDP.

The Threadripper 2920X is the entry point into the second-generation family of chips. This "modest" processor has 12 cores and 24 threads, a base frequency of 3.5 GHz (the same as the 2950X), and a single-core boost clock of 4.3 GHz. The L3 cache amount is fixed at 32 MB, and users should use a cooling solution that can deal with a 180-W TDP.

AMD says the 24-core Ryzen Threadripper 2970WX will sell for $1,299 while the smaller 12-core Threadripper 2920X will land for a relatively affordable $649. The chips will be available on October 29.

Along with the news on the oven-fresh chips, AMD's talking up its Dynamic Local Mode software. The upcoming package is a Windows 10 service that's meant to boost application performance on the four-die 2970WX and 2990WX CPUs by keeping demanding threads on the chips' two dies that can directly access system memory.

The company has a blog post dedicated to the subject with all the details and a benchmark showcase displaying potential performance boosts in a number of games. For the record, although the feature is limited to the Threadripper 2990WX and 2970WX, it'll work fine on systems with X399 motherboards. Dynamic Local Mode will appear on October 29.

Comments closed
    • Johnny Rotten
    • 1 year ago

    So I’ve only been paying half an eye to AMD the last couple of years and i could be way off base so please correct me if I’m wrong…

    1) even with these new threadripper cpus, intel is still the performance leader when it comes to gaming (which is basically what I use my computer for 80% of the time) T/F

    2) threadripper is still significantly more power hungry/hotter than intel cpu’s T/F

    I’m trying to figure out if I want an AMD cpu for my next home pc but as far as I can tell I still want an intel cpu…

      • blastdoor
      • 1 year ago

      If you are primarily interested in gaming, go Intel.

        • JustAnEngineer
        • 1 year ago

        Threadripper is [b<]not for gaming[/b<]. Games are not using 32 (or 64) threads. If gaming is your primary use, you should be shopping for Core i7-9700K at 8 cores / 8 threads or Ryzen 7 2700X at 8 cores / 16 threads or Core i7-8700K or Ryzen 5 2600X (both at 6 cores / 12 threads). Rather than spending hundreds of dollars more on a workstation CPU for your gaming rig, you should invest in a faster graphics card.

      • Waco
      • 1 year ago

      2). It depends on the workload but in general TR/Ryzen is on par or slightly ahead in task energy when considering multi-threaded tasks.

    • DarkUltra
    • 1 year ago

    For some reason I can’t scroll the table yo see prices on my Android 8.0 phone with Chrome

    • ptsant
    • 1 year ago

    Isn’t linux already implementing something like Dynamic Local mode? I know that the kernel is (can be made?) NUMA aware, but don’t remember if there are particular settings to modify and TR is already recognized.

    • enixenigma
    • 1 year ago

    The link to the blog post is broken.

      • derFunkenstein
      • 1 year ago

      Here you go:

      [url<]https://community.amd.com/community/gaming/blog/2018/10/05/previewing-dynamic-local-mode-for-the-amd-ryzen-threadripper-wx-series-processors[/url<]

      • morphine
      • 1 year ago

      Link fixed. It was a long day.

    • willyolioleo
    • 1 year ago

    Awesome that the feature will work on x399. I wonder when the last generation of TR chips will be that work on that chipset…

    • techguy
    • 1 year ago

    Dynamic local mode is exactly what the 4-die Threadripper SKUs need to deliver performance consistency across the highest number of workloads. Very good decision AMD. That 2970x is a killer price too.

    • Chrispy_
    • 1 year ago

    My experience with the 2990WX left me impressed at just how much power AMD can cram into a single, readily-available consumer motherboard but the awkward division of memory channels and forced NUMA-only mode means that anything more than a dual-die Threadripper runs into an increasingly long list of caveats for them to make sense.

    The 2950X is definitely more appealing; 4.4GHz and true quad-channel memory with no caveats is almost certainly the most suitable choice for a great many people wanting to move beyond S1151 or AM4.

    We have an EPYC server (2x EPYC 7401) and despite the lower clockspeeds, it’s still quicker than the 2990WX. Sure, a dual EPYC 7401 workstation will probably cost you another $2000 over an equivalent 2990WX build, but if your workload involves datasets that are memory-intensive, the Threadrippers beyond the dual-die 2950X are bottlenecked by that 25GB/s interface between dies and near-doubled memory latency.

    If your workload fits in cache and scales exceptionally well, you are the niche market that the 2990/70 models are best suited for. The other niche market (that should never be underestimated) is rich idiots that have to have the most expensive thing no matter how (un)suitable they may be πŸ˜‰

      • Waco
      • 1 year ago

      Dual 7401s versus a 2990WX isn’t a very fair fight…16 memory channels versus 4. πŸ™‚

        • blastdoor
        • 1 year ago

        I wonder how the 2990wx would perform with HBM

      • Voldenuit
      • 1 year ago

      [quote<]My experience with the 2990WX left me impressed at just how much power AMD can cram into a single, readily-available consumer motherboard but the awkward division of memory channels and forced NUMA-only mode means that anything more than a dual-die Threadripper runs into an increasingly long list of caveats for them to make sense.[/quote<] Yeah, my first thought when I saw the block diagrams was, 'WTF, there are CCXes that have no connection to memory???' It's definitely fine for some workloads, but I can definitely understand why some other workloads balk on Threadripper.

      • freebird
      • 1 year ago

      Yeah, it a little sad, we have Production Servers bought built mid-2015 one of which would get owned by my PC build (Ryzen 1700 & 64GB) and the rest would have trouble out performing a 2950x build with 128GB.

      • blastdoor
      • 1 year ago

      I have a 2990wx system running Linux and am still in the process of figuring out how to get the most performance out of it for my workloads.

      One thing I’m curious to learn more about is a daemon called numad (http://manpages.ubuntu.com/manpages/xenial/man8/numad.8.html).

      Allegedly this daemon monitors running tasks and attempts to efficiently balance loads across processors. Does anyone have experience with this?

        • Chrispy_
        • 1 year ago

        Sorry, no. Our machines with high core counts are either hosting mulitple VMs, or raytracing & environmental simulations.

        The thing with TR2 is that you don’t [i<]want[/i<] to balance loads across processors. You want to run it in NUMA mode and only load up the higher-latency cores when you need compute performance with a highly-scalable workload. In that situation, even a hampered core with low bandwidth and high latency is an improvement over nothing. Edit - To clarify, don't think of the 2990 as a 32-core CPU. It technically does have 32-cores but only 16 of them are comparable to a Ryzen cores. The bonus cores add somewhere between -20% and +75% performance, depending on whether your OS, drivers, and applications know what to do with them, and this is where the Ryzen Master software and Windows have an advantage.

          • blastdoor
          • 1 year ago

          I think it’s accurate to say that those extra 16 cores are “bonus cores” that add between 20 and 75% performance.

          But the Windows part — I’m not so sure about that. I went with Linux on my 2990wx system because I’m very skeptical regarding the ability of the Windows task scheduler to walk and chew gum at the same time let alone handle the eccentricities of the 2990wx. I’m not religious about this stuff, though — if Windows is a better option than Linux for the 2990wx, I’ll switch.

          • Waco
          • 1 year ago

          “Bonus cores” == “cores you have to try slightly harder then normal to utilize”

          They are no different than the Intel cores that are “far” from the memory controller on the mesh or ring bus. If you can tailor your workload to understand NUMA, you’re good to go.

          The 1950X in my NAS even required a bit of tweaking to maximize performance. Default configuration was ~25 GB/s memory bandwidth and bad transcoding performance. Naive NUMA config yielded 65 GB/s memory bandwidth and 3x 4K transcodes on the fly.

            • psuedonymous
            • 1 year ago

            [quote<]They are no different than the Intel cores that are "far" from the memory controller on the mesh or ring bus. [/quote<] [url=https://www.pcper.com/reviews/Processors/Intel-Core-i9-7900X-10-core-Skylake-X-Processor-Review/Thread-Thread-Latency-and-<]There is no 'near' or 'far' cores on the mesh bus, they all have equal access latency[/url<].

            • Waco
            • 1 year ago

            I wish that were true on the larger chips.

    • freebird
    • 1 year ago

    I think the Base & Boost Clock columns are reversed… unless 4.2GHz for 32 cores is a new 7nm part. ; D

      • morphine
      • 1 year ago

      Oops. Sorry, I shouldn’t wish for so much. Fixed.

    • Krogoth
    • 1 year ago

    Lisa Su: Now witness, the computing power of this fully operational workstation CPU line-up!

    #PoorSkylake-X
    #PoorCasadeLake
    #Poor10nm

      • freebird
      • 1 year ago

      You forgot one after CascadeLake and before 10nm

      #Poor PooperLake

      Edit:
      Oops, I guess people are upset I got the code name wrong, it is Cooper Lake not Pooper Lake, but it probably will stink up the Lake series.

      • Srsly_Bro
      • 1 year ago

      Bro, rumor is AMD Rome was to compete against Ice Lake.

      Let me help you with some additions:

      #PoorIceLake
      #RIP10nmHello12nm
      #PoorIntelFabs
      #MeToo@CEO

      The company can’t get chips out the door or keep the CEO’s business in his pants. The end is near for Intel, culturally and financially.

        • freebird
        • 1 year ago

        Actually, it wasn’t a rumor it was a quote from AMD’s Forrest Norrod near the end of this article:

        “Rome was designed to compete favorably with β€œIce Lake” Xeons, but it is not going to be competing against that chip. We are incredibly excited, and it is all coming together at one point. ”

        [url<]https://www.nextplatform.com/2018/06/20/amds-epyc-return-to-the-datacenter-ring/[/url<] The whole article is a pretty good read on what AMD was/is planning for the server space. Edit: Interesting 2nd quote in the article: "The AMD plan all along was for the Rome Epyc chip to come to market somewhere between 20 months and 24 months after the Naples Epyc chip, and then to get on a cadence of new chip launches every 12 months to 15 months under the assumption that Intel would crank out new chips like crazy and try to leverage its manufacturing prowess." Epyc (Naples) was officially released on June 20, 2017, so that mean 7nm Epyc should be expected in Q1/Q2 of 2019. That is well before Cooper or Ice Lake and Milan (7nm+) should be out to combat Ice Lake in late 2020.

        • Kretschmer
        • 1 year ago

        The end is near for Intel financially? Let’s check their Q3 results
        Intel: $17B Rev/$5B Inc
        AMD: $1.7B Rev/$0.1B Inc

        Welcome to reality, here are your cold hard numbers.

          • Srsly_Bro
          • 1 year ago

          I’m going to bookmark this page and if you’re around next year, we’ll see who’s right.

            • TheRazorsEdge
            • 1 year ago

            I can’t tell if you’re trolling or just ignorant.

            Let’s forget about each company’s cash reserves (although Intel has a lot, and AMD has… debt). AMD doesn’t have the capacity to replace Intel.

            So go ahead and bookmark that comment, then maybe learn a lesson in a year’s time.

      • tsk
      • 1 year ago

      Where is chuckula, what did you do to him?

        • jihadjoe
        • 1 year ago

        He’s at Krogoth’s house, using Krogoth’s PC?

Pin It on Pinterest

Share This