Rumor: Leaked AMD roadmaps give a peek at future CPUs

The folks over at Videocardz got their hands on what appear to be some decidedly non-video-card-related internal AMD roadmaps. The slides describe AMD's enterprise CPU and APU products, but given that they are dated February 2016, they could be quite out of date. Still, if accurate, the presentation gives us an interesting picture of AMD's future plans. Just ring up your local supermarket and ask for a truckload of salt before proceeding.

Source: Videocardz.com

The purported roadmaps list a number of products that have launched and a great many more that have not. Perhaps the most salient hypothetical news is that AMD is planning a 48-core-per-socket CPU fabricated on a 7-nm process. According to the roadmaps, this would-be processor was code-named "Starship." Starship would be the next Opteron at the top of AMD's product stack after the 32-core "Naples" processor that is expected to launch any day now. 

If the slide deck is real, AMD could be expecting to launch Starship as soon as next year—although, once again, remember that these roadmaps would be fourteen months old. Shortly after Starship we could see "Grey Hawk" and "River Hawk" CPUs also fabricated on a 7-nm process, but those products seem highly speculative, seeing as they're listed for a hypothetical 2019 release.

Source: Videocardz.com

This new-old rumor also seems to lend some credence to the existence of the all-but-known 16-core Zen processors. The slides refer to the chips as "Snowy Owl" and curiously claim that they would only be offered with BGA packaging. More recent rumors have identified the upcoming processors as LGA packages akin to Naples. In any case, the slides explain that "Snowy Owl" uses one or two of the "Zeppelin" dies found in current Ryzen processors to scale between four and 16 Zen cores.

Further down the slides, we find some possible information about AMD's upcoming APUs. The slides talk about AMD's embedded processors, which means the information may not translate directly to consumer products. Still, the data in the rumors is tantalizing. One slide describes "Great Horned Owl" as an APU featuring four Zen cores, 11 GCN CUs, support for VP9 video decoding at 4K resolution and 60 Hz, and support for up to four simultaneous 4K displays. 

The slides describe the graphics bits of those purported APUs as being based on the GFX9 architecture—that's Vega, for those unaware—and go on to say that they can also decode and encode 10-bit H.265 video. The would-be chips have lots of connectivity, too: four USB 3.1 connections, four SATA ports, dual Gigabit Ethernet connections, and the usual 16 lanes of PCIe 3.0 connectivity. If true, AMD's presentation says that the chips will go into BGA format as well as AM4 sockets, and that TDPs will range from 12W to 65W.

There's a lot more information in the slides, and though they may be outdated, they're still interesting to look at. Just keep the salt handy, because even the slides themselves say that the information is subject to change.

Comments closed
    • ronch
    • 3 years ago

    Would be interesting to see what Ryzen on 7nm can do.

    • ptsant
    • 3 years ago

    The Naples product has been clear for a very long time but is not very relevant for consumers. What is becoming clearer is the development of a HEDT line with 12-16 cores and 8 channels. That is really cool and will completely destroy the Intel HEDT line, if true. We’ll have to wait and see, but I think Skylake-X and Kaby-X will have strong competition.

      • Krogoth
      • 3 years ago

      Kaby-X is already DOA if the one chart is true. Lower-end Skylake-X stuff already has to contest with the current Ryzen R7 line-up. If AMD manages to make a HEDT version of Naples platform then those high-end Skylake parts will have something to fight against.

        • chuckula
        • 3 years ago

        Kaby-X is there 1. To be a cheap entry point into the HEDT platform; and 2. To be a strong overclocker that performs well in games.

        Slapping 2 RyZen dies onto an interposer is not going to be cheap (even if it undercuts the bigger Skylake X parts) and I’m having a hard time seeing two dies in that configuration overclocking better than a regular RyZen chip.

      • chuckula
      • 3 years ago

      12 – 16 cores and 4 memory channels that is. There’s only 2 working dies on those parts, although they apparently use the same socket as the full-sized Naples server for economic reasons.

        • ptsant
        • 3 years ago

        Yes, you are right about the memory. I somewhat hastily confused the channels with the DRAM slots on the motherboard.

      • ImSpartacus
      • 3 years ago

      With Turbo Boost 3.0 pushing Skylake-X’s clocks to 4.5 GHz, it looks like Intel isn’t prepared to be losing any lightly threaded benchmarks for the time being.

    • Unknown-Error
    • 3 years ago

    Starship?

      • Klimax
      • 3 years ago

      To get to Polaris on Starship using Vega rocket.

    • Dahak
    • 3 years ago

    AMD now launching the Dreadnought class starship HMS Threadripperâ„¢.

      • KeillRandor
      • 3 years ago

      HMS? I thought AMD was an American company? 😛

      • Pancake
      • 3 years ago

      A proper British name would be Thready McThreadface.

        • YellaChicken
        • 3 years ago

        Hahaha, thanks m8, I’ll be laughing all day at that one.

    • Kougar
    • 3 years ago

    AMD had better be playing Starship – “We built this [s<]City[/s<] CPU" at the launch of that 7nm 48/96 core monster.

      • Chrispy_
      • 3 years ago

      As long as it’s that and not Starships by Nicki Minaj, I’m with you.

      • mganai
      • 3 years ago

      Nothing’s gonna stop us, nooooooooowwwwwwwww!!!!!

    • Chrispy_
    • 3 years ago

    Of very dubious accuracy and so old that they’re probably irrelevant.

    According to those charts we had 32-core server processors in 2H 2016 (9-12 months ago) and the consumer Zen would be a fall 2016 native quad core part, and Zeppelin is the code-name for AMD’s Zen-based Opteron.

    Timing wrong.
    Product wrong.
    Naming wrong.
    Wrong wrong.

    It’s all just wrong.

      • RAGEPRO
      • 3 years ago

      Err, well, that first graph seems to indicate a launch window between Q3 2016 and Q2 2017, which is where we’re at now. Zeppelin (if you check the rest of the slides) appears to actually be the codename for the 8C Zen die used in Summit Ridge; Snowy Owl uses 1 or 2 Zeppelin dies.

      I agree that they’re old but they seem to be fairly accurate.

        • Chrispy_
        • 3 years ago

        Hmmm, I guess that would make it less [i<]in[/i<]accurate, simply because the Snowy Owl/Naples product seems to have a couple of months left to run on that chart - giving some credence to why nobody has even caught a whiff of these chips yet. Even so, Snowy Owl is dual ZP in a single socket, Naples is quad ZP in a single socket. I'm certain we'd have seen leaks/rumours of these 16c/32t and 32c/64t Opterons by now if they were even remotely close (like, within the next 12 months) Getting these out to market before the end of Q2 2017 is one hell of a stretch and impressively unlike AMD. Perhaps I should say "incredibly unlikely" and the [i<]literal[/i<] meaning of incredible at that. No; Honestly? I'm far more inclined to believe my original assumption of [i<]these being production windows[/i<] - simply because the Slides are from 2016 and the charts portray the historical production windows and cutoff dates of Excavator production runs that ended in 2015. There's no "launch window" uncertainty for stuff that already happened in the past.

          • RAGEPRO
          • 3 years ago

          [url=https://techreport.com/news/31549/amd-naples-platform-prepares-to-take-zen-into-the-datacenter<]Um?[/url<]

            • Chrispy_
            • 3 years ago

            OMG, how did I forget about that?

      • Antimatter
      • 3 years ago

      I think the right edge of the rectangle indicates the launch date. The ARM 1100 CPU launched at the beginning of last year and Naples will be launching in middle of this year, matching the slide. The enterprise APU will likely launch in Q4 this year a few months after the consumer products. That puts the 7nm Starship chip firmly in early in 2019 not 2018, which seems more likely.

      • ImSpartacus
      • 3 years ago

      There’s a disclaimer at the bottom right of the first slide saying the placement of boxes will not match release pace.

      TR’s copy is too low res, but the original makes it very clear.

        • Chrispy_
        • 3 years ago

        That solves the problem of me trying to work out why the box placement didn’t match up with historical product releases 🙂

    • Omniman
    • 3 years ago

    Interesting to see a life expectancy listed on the “Great Horned Owl”.

      • Wirko
      • 3 years ago

      Ten years? They could as well say “forever”.

        • synthtel2
        • 3 years ago

        If it’s 10 years of 24/7 90C load on the highest-clocked parts, they could as well say forever (but it would still be useful data for overclockers). Those other conditions are important, though.

    • chuckula
    • 3 years ago

    While there’s a lot to discuss in the slides, the most unexpected thing is on the last slide for the APU that claims the GPU is using [s<]"GCN"[/s<] [strike that, it's literally "CGN", is that the new Vega nomenclature or just a typo in a slide??] units. There was at least a rumor that these APUs were using the same GPU architecture as Vega and hasn't Vega got a new name for its compute units?

      • Fursdon
      • 3 years ago

      Per the recent Linux Raven support added :

      [url=https://lists.freedesktop.org/archives/amd-gfx/2017-May/008916.html<]1[/url<] [url=https://lists.freedesktop.org/archives/amd-gfx/2017-May/008934.html<]2[/url<] [url=https://lists.freedesktop.org/archives/amd-gfx/2017-May/008944.html<]3[/url<]

        • chuckula
        • 3 years ago

        So it’s being grouped with Vega. We’ll see more later.

      • Chrispy_
      • 3 years ago

      Vega’s new architecture is called NCU.

      • ptsant
      • 3 years ago

      I don’t know whether they are keeping the familiar GCN name for the VEGA cores, but I would expect the APUs to feature Polaris cores. I might be mistaken, but it would be relatively hard to develop both at the same time. I mean, Vega is not yet released and the APUs are probably taped out or close to that. Anyway, if indeed they jumped directly to Vega cores, that is very good news for people planning on buying an APU.

      • ImSpartacus
      • 3 years ago

      The initial Polaris-Vega-Navi roadmap had the memory tech for Navi listed as “NexGen”. People were wondering if that was some new forthcoming memory type to replace HBM.

      Nope, just a typo that was corrected to “NextGen” (AMD-speak for “something new”).

      So I think this is just a typo.

Pin It on Pinterest

Share This