Leaked slides detail 28-nm AMD ‘Deccan’ APU

Although most of the PC hardware industry seems to be on vacation right now, the rumor mill keeps churning away. Turkish site Donanim Haber has published yet another batch of official-looking AMD slides, this time detailing plans for upcoming "Wichita" and "Krishna" APUs. Due in 2012, the 28-nm chips will be a part of the "Deccan" platform. Interestingly, Deccan doesn’t appear to include any other silicon—it looks like the chipset, or Fusion Controller Hub in AMD’s parlance, will be brought into the APU.

According to the slide, Deccan’s FCH will include Serial ATA and USB connectivity. The speed of the former isn’t provided, but USB ports will be available in 2.0 and 3.0 flavors. Four PCI Express lanes provide a conduit for additional peripherals.

Deccan’s CPU component will purportedly be based on the Bobcat cores found in existing Zacate and Ontario APUs. Each core will be 20% faster than what’s found in Ontario, and up to four of ’em will take up residence on a single APU. Turbo Core is part of the equation, as well, and it looks like Deccan will balance the clock speeds of its CPU and GPU components based on the application load. Incidentally, that GPU component is claimed to be 25% faster than Ontario’s IGP.

One of the slides mentions "lower TDPs targeted for new form-factors," suggesting Deccan derivatives could be aimed at tablets rather than netbooks and ultraportable notebooks. Another shows 8W and 20W Deccan parts, which puts the new APU slightly above and below the TDP ratings of Zacate and Ontario APUs, which are pegged at 18W and 9W, respectively.

Comments closed
    • DavidC1
    • 8 years ago

    Watt-to-Watt comparison of CPUs and I/O chips don’t work. CPUs reach TDP much more frequently in more situations than I/O chips do. Going from 18W CPU + 5W I/O to 20W combined is unforunately, not lower at all. In order for the I/O chip to reach anywhere near 5W TDP, it needs to have every port available fully utilized simultaneously. It’s not that common.

    They’ve been saying “Enhanced Bobcat” cores, which may mean as little as no more than a die shrink. If anything, the gain is coming from the clock speed. I think what I said earlier works perfectly. 20% faster than Ontario for CPU, with 2x cores, mean 1.6GHz 4 cores. Double the core with same power is what Moore’s Law is.

    • ronch
    • 8 years ago

    Sounds like an enticing chip for laptops, if not Ultrabooks. Is that 20% increase in performance due to clock speed increases or architectural enhancements?

    • HighTech4US2
    • 8 years ago

    > Four PCI Express lanes provide a conduit for additional peripherals.

    Deccan looks to be useful only as an Intel Atom in performance (in 2012) and no possibility of using a modern GPU (which needs 8 to 16 lanes) on it.

    And the power of 8 to 20 watts is way too high.

      • Hattig
      • 8 years ago

      It integrates a modern GPU. Sure, not a powerful GPU, but totally modern.

      If you want to do PC gaming, then you need something a bit more powerful than an embedded/nettop/netbook/tablet.

      Bobcat cores make Atom look anaemic already.

      Yuba, the integrated I/O on Wichita, is specced for tablets and netbooks – one SATA, some USB2 and USB1, essential I/O. The PCIe is used to connect to Hudson 2, the I/O chip for bigger systems that need more SATA, USB3, and so on.

      You can still run a reasonable GPU through PCIe x4 however, but why would you want to add such a power hungry thing to a low power platform in the first place?

      • ET3D
      • 8 years ago

      Modern GPU’s function pretty decently with 4 lanes. On the other hand most games are CPU bound on the E-350 and should remain so with the next gen. Upshot is that there’s little need for more than four lanes.

      • Goty
      • 8 years ago

      [quote<]And the power of 8 to 20 watts is way too high.[/quote<] For what? My X370 gets nine hours of useful battery life out of its ~23W combination of E-350 and Hudson M1 FCH. I won't complain about getting roughly twice the CPU resources and a small bump in GPU performance for a ~15% reduction in power consumption.

      • NeelyCam
      • 8 years ago

      Successful troll – good job!

    • NeelyCam
    • 8 years ago

    Here’s my take (and I bet everybody’s been waiting for my take on a new AMD product).

    The key pieces of information:

    [b<]1) Cores being 20% faster than [u]Ontario[/i][/b<] Ontario is running at 1GHz (dual-core) or 1.2GHz (single-core). Being 20% faster than an Ontario core means it's still 10-25% slower than a Zacate core. [b<]2) 50% more Giggaflopps (i.e., 1/1.5=0.6667=0.7= 30% more "Compute benefits")[/b<] This is in line with the calculation above. 4 Deccan cores, each 0.75x as fast as Zacate = 1.5x more than a dual-core Zacate. So, Deccan cores look 25% slower than Zacate cores (=1.2GHz). [b<]3) 18W+5W (DC Zacate + Fusion Controller Hub) -> 20W (QC Deccan w/ integrated hub)[/b<] Instead of going for more single-thread performance with 28nm, AMD went with lower power consumption and better graphics (like the graphics weren't good enough for this CPU class already)... [b<]4) Turbo!![/b<] ... but here's the thing that addresses the single-thread performance.. with Turbo, maybe they can still keep the single-thread performance at the Zacate level (or, who knows, maybe even higher... maybe the 28nm chip could clock up to 2GHz) So more multithreaded horsepower, hopefully more single-threaded horsepower (on a temporary basis), and lower power consumption. I think this could be a compelling product, fixing the shortcomings of Brazos. A cheaper alternative to higher-performance, same-power-consumption IvyBridge Ultrabook chips. I couldn't recommend Zacate to any of my friends, but I could recommend this - I think it will meet my requirements of what is "good enough" (assuming the Turbo is good).

      • DavidC1
      • 8 years ago

      One thing I’d like to correct. The slide has 80GFlops for the 18W Brazos platform and 120GFlops for the Deccan one.

      18W Brazos 80GFlops: HD5450 core @ 500MHz
      20W Deccan 120GFlops: HD5450 core @ 750MHz

      Now that’s a 50% gain, but there’s also the Bobcat refresh
      E-450 18W, 90-96GFlops: HD5450 core @ 560-600MHz

      That’s how they might have got the up to 30% gain for graphics.

      20% faster than Ontario’s CPU. There’s the C-50 with 1.0GHz clock and refresh C-60 with the 1.0GHz clock and 1.33GHz Turbo. That may mean.

      Deccan 20W: 4 core @ 1.2x C-60 Turbo = 1.6GHz

      To summarize:

      Deccan 20W: 4 core @ 1.6GHz with Turbo, and HD5450 core @ 750MHz

    • sschaem
    • 8 years ago

    The slide state that Decan get 60GFlop vs 40 for Brazos.
    Exact same ratio with Brazos and Deccan, +50% raw compute.
    But then the slide mention upto 30%… what did I miss.
    And 50% cant come from adding two core, that would represent 100%
    So my guess is that 50% come from double the shader count.
    50% more shader compute power, and 20% CPU integer.

      • rechicero
      • 8 years ago

      Maybe they meant Brazos get 30% less GFlop than Deccan? 40 is 30% less than 60. 60 is 50% more than 40. Percentiles are funny 😛

    • d0g_p00p
    • 8 years ago

    Interesting, so this will not have a south bridge chip at all? So it’s really a x86 system on a chip.

    • bcronce
    • 8 years ago

    I can’t wait for a nice dual module BD based chip to use as a pfsense firewall. My 3700 will have to last a bit longer. So excited about these new low power CPUs AMD is working on.

      • Farting Bob
      • 8 years ago

      Do you need a BD based chip to run a firewall? Atom can do that fine right now..

        • shank15217
        • 8 years ago

        It all depends on the firewall…

    • Goty
    • 8 years ago

    [quote<]Another shows 8W and 20W Deccan parts, which puts the new APU slightly above and below the TDP ratings of Zacate and Ontario APUs, which are pegged at 18W and 9W, respectively.[/quote<] Add roughly 5W to the numbers for Zacate and Ontario for the Hudson M1 FCH and those power numbers look pretty good.

    • OneArmedScissor
    • 8 years ago

    I’m curious what integrated SATA does to storage latency more so than cost and computer design. It would be nice if that even shaved a second or two off of boot/sleep/load times.

    They could probably make a pretty whacky server platform out of this. One chip for every hard drive! I have no idea why anyone would do that, but there have already been some pretty outlandish uses of Atom, and it doesn’t even have a true IMC yet.

      • willmore
      • 8 years ago

      Likely nothing. It’s probably architecturally just how things are now. There’s probably even an internal PCI-E link. The latency between the CPU and peripherial is pretty darn tiny compated to the latencies inside th storage device. Amdahl’s law says it’s silly to mess with it, then.

    • Arclight
    • 8 years ago

    [quote<]Leaked slides detail 28-nm AMD 'Deccan' APU[/quote<] Wait, wasn't the next die shrink to 22nm?

      • OneArmedScissor
      • 8 years ago

      Bobcat is made at TSMC and follows GPU manufacturing cycles, at least for now. Most of the chip is a GPU and what used to be the north and southbridges, and those were always TSMC’s thing.

      • bcronce
      • 8 years ago

      28nm is a half-node of 32nm and 32nm as a huge problem for companies. Intel is the only company with a working 22nm.

    • odizzido
    • 8 years ago

    My ontario laptop is pretty good and is way better than the atom alternative for me. The only problem I have with it is it seems to have really high power draw at idle. If they can do better on that front this could be a great chip.

      • NeelyCam
      • 8 years ago

      Hard to believe… OS software issue?

        • wierdo
        • 8 years ago

        Yeah probably, supposed to be pretty power efficient according to reviews.

        On the other hand it’s also possible that the laptop came with a smaller battery to save on costs, seen some like that.

        • odizzido
        • 8 years ago

        Don’t think it’s a software issue. I always idle at 0-5% CPU use on windows 7. By really high I mean relative to atom.

        I don’t think it’s as good as atom idle because when doing nothing my c-50 laptop feels a fair bit warmer than my atom one. Yes I know the heatsink/layout could be the cause, but I still think it’s less efficient by a fair amount.

        I am not saying it’s horrible or anything, it’s fairly decent. I just feel that they can probably improve things in the idle department.

          • cegras
          • 8 years ago

          Is battery life the same? Heat is not temperature, and certainly not a direct indicator of power draw.

            • NeelyCam
            • 8 years ago

            This. Heat is a better indicator of a piss-poor cooling solution.

            • willmore
            • 8 years ago

            If the APU can take the heat, it’s more power efficient to run the fan slower (or stay passive) and let the processor cook.

            • NeelyCam
            • 8 years ago

            Those tiny fans consume next to nothing. Heat increases transistor leakage because it reduces the threshold voltage.

            But I think your point is more or less the same as mine – heat doesn’t necessarily mean poor power efficiency

      • Hattig
      • 8 years ago

      It will – it adds C6 power states (i.e., power gating) on a per-core basis.

      It can idle at 1W now, including the chipset, graphics, four cores, etc. 3.5W to play back BluRay too.

      • willmore
      • 8 years ago

      Did you try BrazosTweaker on it, yet? You can play with the voltages and the clock divider for the different P states. Saved a few watts on my C-50 laptop.

    • abw
    • 8 years ago

    So they promise 20% better IPC than the current bobcat…

    This would bring it ahead of a K8 CPU in integer.
    In FP , it will be still below..

      • Goty
      • 8 years ago

      Most common desktop workloads are int-heavy, anyhow, so that should still be alright.

        • abw
        • 8 years ago

        Agree , theses cats are not for renderings and the likes…
        People using zacate xxxbooks seems happy with them , indeed…

      • NeelyCam
      • 8 years ago

      No IPC change. See my post with the calculations.

        • abw
        • 8 years ago

        As usual , your calculations have a serious flaw that render the whole thing useless…

        The 50% + Gflops figure is the GPU capability when using OpenCL, it has nothing to
        do with the X87/SSE FP capabilities of the cores, so your estimation of the new
        cores of being 75% of a Zacate is also built on a house card and as a result,
        all your post is futile…

        Isnt that weird.??….Life s not always fun…..

          • NeelyCam
          • 8 years ago

          “As usual”? NeelyCam 4 – AMD fanbois 0.

          Life being fun is mostly a state of mind.

            • willmore
            • 8 years ago

            “NeelyCam -4”

            There, fixed that for you.

            • NeelyCam
            • 8 years ago

            That was half-clever

        • Hattig
        • 8 years ago

        The cores are meant to be enhanced Bobcat cores, from previous AMD slides. Presumably there is an IPC change, but more likely instruction set support enhancements.

        I wouldn’t make claims about the clock speeds just yet though. There’s a lot still unknown.

      • maroon1
      • 8 years ago

      The slides says that it will be to 20% faster

      “up to” usually means best case scenario or cherry-picked benchmarks. It not the average performance

      Moreover, the slide doesn’t say 20% better IPC. The 20% improvement could be as result of higher clock speed.

        • abw
        • 8 years ago

        “Enhanced CPU cores” means just higher clock speed for you ?…

        [url<]http://www.laptopspirit.fr/wp-content/uploads/new/amd-roadmap-2012-2.jpg[/url<]

Pin It on Pinterest

Share This