Rumor: Vega Frontier Edition board power ratings surface

Unless you happened to miss yesterday's reveal, all the talk in the graphics card world right now is about AMD's Vega Frontier Edition graphics cards. The price reveal is interesting enough if it's accurate, but there's another tidbit of as-yet-unofficial info that found its way into the wild—total board power ratings, courtesy of pro graphics retailer Exxact. Steal your budgie's salt lick and read on.

First, the exact rumored figures. The air-cooled standard-issue Vega FE card with a blower cooler is rated for a square 300W of "max board TDP." Meanwhile, its under-water cousin can pull a whopping 375W at peak. These numbers may look a little high at first, but let's try and keep some perspective. The performance information presented is also for the air-cooled Vega card only, so the liquid-cooled version could offer even higher performance to go with its power draw.

While it's tempting to draw conclusions about the performance per watt of this card from these figures, there are a few things to keep in mind. First and foremost, the air-cooled Vega FE isn't really meant to be compared to standard-issue GeForces. We think the natural competitor for this compute-focused card is the Quadro GP100. That card already pulls a respectable 235W of power, or about 28% less than the rumored 300W figure for the air-cooled Vega FE. In turn, we spitballed that the GP100 would deliver about 26% lower peak performance in both FP16 and FP32 operation than the Vega chip on board the Frontier Edition. If those projections hold, at least the FLOPS-per-watt increase would seem to be linear.

Second, Vega is a new architecture, and drawing direct comparison to its predecessors is an endeavor doomed to failure. This may be stating the obvious, but far too often we witness megahertz and mega-watts comparisons between apples and coconuts. We don't know full implementation details of the Vega chip on board the Frontier Edition card yet, but we do know that several new things are going on with this architecture compared to older Radeons, like the High-Bandwidth Cache Controller. Those fancy features may require a lot of power to operate.

Third, high-performance cards simply drink up juice like it's going out of style, and there's no word how far AMD cranked the clock knobs on the Vega FE (though we're guessing it's around 1600 MHz at peak). Heck, not that long ago, the R9 290X's board power was 290W. It also pays to think of voltage and frequency scaling as an exponential function, because it is. You need to look no further than the RX 580 and realize that its power draw is roughly 40W over the RX 480 for a relatively minor increase in clock speeds. We should be able to find out more soon, since the Vega Frontier Edition launches June 27.

Comments closed
    • Bumper
    • 2 years ago

    We’ll see. For what it’s worth raja koduri has said his first full arch with RTG will be Navi. I expect RTG to simply hold the line with VEGA in the discrete graphics market.

    I am looking forward to the Zen and VEGA apu.

      • ImSpartacus
      • 2 years ago

      I doubt that Navi will be fundamentally different. It’ll probably allow for initial MCM efforts, but that’s probably it.

      And don’t get your hopes up concerning Raven Ridge. It’ll excel in laptops, but won’t really move the dial in larger TDP ranges.

        • Bumper
        • 2 years ago

        I should have been clearer. My hopes are in the laptops. I plan on buying a new one in the fall.. 🙂

        • the
        • 2 years ago

        The MCM scaling though will be interesting since the only limitation will be power consumption. AMD’s eating the cost for interposers and going MCM on top of HBM doesn’t add much in that regard. AMD could put together enough aggregate to make the GV100 chip look small in comparison. This also means that AMD shouldn’t be operating outside of the optimal voltage/frequency curve.

        How well Raven Ridge performs would be dependent upon the existence of a model with HBM/WideIO. That would be a game changer in the market.

    • ultima_trev
    • 2 years ago

    I want AMD to do well (not just because I am a Ryzen fanboi) but this doesn’t bode well for them.

    It might be able to outperform a Titan X Pascal that is quite obviously thermal throttled by a crappy founder’s edition cooler, but what about a custom cooler 1080 Ti that will boost to 1900+ out of the box (thus generating 13.6+ TFLOPs compute)?

    Nevermind the fact that Radeons always under performed relative to their compute throughput…

    • jihadjoe
    • 2 years ago

    AFAIK the gimped performance on Titan Xp is all down to it not having a pro driver. A Quadro with a much smaller GPU has higher performance than both it and Vega FE in solidworks, so hopefully this sort of pressure goads Nvidia into writing a driver that enables pro features for Titan.

    • ronch
    • 2 years ago

    Oh well, 4 days to go. If the power efficiency isn’t there I hope the performance is at least comparable to anything from Jensen’s gang.

    • Unknown-Error
    • 2 years ago

    Wait! If this is NOT a gaming card, then why are they comparing it to Titan XP? Even the $5000 Quadro P6000 has a TDP of 250W. P4000 is only rated at 105W.

      • gc9
      • 2 years ago

      Recall where the Titan X(p) was announced: At a deep-learning algorithm conference, at a ceremony where the first one was awarded to Baidu Chief Scientist Andrew Ng.

      • NoOne ButMe
      • 2 years ago

      It looks bad. Really bad. And if true, it will be really bad.

      But “max board power” is not “typical board power” that AMD uses.

        • ImSpartacus
        • 2 years ago

        Are you suggesting that the air-cooled and water-cooled versions use different boards with different power delivery capabilities?

        I doubt that they use different boards.

          • NoOne ButMe
          • 2 years ago

          no, but max power draw would be related to the cooling capability as well…. If you can only cool 300W, setting the max power to 375W does not make very sense. yes.

    • Srsly_Bro
    • 2 years ago

    1600 MHz doesn’t come cheap on 14 NM lpp.

    • willyolioleo
    • 2 years ago

    Why is half precision also fp32?

      • ImSpartacus
      • 2 years ago

      Typo. This isn’t a world-class vendor.

      • ronch
      • 2 years ago

      It’s complicated.

    • Tristan
    • 2 years ago

    Let they rename it to “Vega 375”

      • ronch
      • 2 years ago

      Oh wow. A joke from Tristan.

        • Redocbew
        • 2 years ago

        Is that what it is?

          • ronch
          • 2 years ago

          <shrugs>

    • Bumper
    • 2 years ago

    I’m interested to see how this (las Vegas) turns out.

    • USAFTW
    • 2 years ago

    DAMN YOU DOUBLE POST!

    • USAFTW
    • 2 years ago

    FWIW, A cut-down GP104 Quadro P4000 is faster in some aspects of the Specviewperf 12.1 test than the Vega scores above. They’re not fooling anybody (almost) by comparing a “gaming” GPU with GeForce drivers with a workstation GPU utilizing AMD’s Pro drivers.
    [url<]https://hothardware.com/reviews/nvidia-quadro-p4000-and-p2000-pro-workstation-gpu-review[/url<] Edit: Interestingly, a Polaris-based WX 7100 isn't far behind either.

      • Voldenuit
      • 2 years ago

      Yep. Professional OpenGL applications are *severely* hampered on consumer cards and drivers, because AMD and nvidia want to segregate their professional (low volume, high margin) and consumer (high volume, low margin) SKUs, despite the cards themselves using identical silicon.

      That said, I’m surprised the Titan X and Xp aren’t “good enough” for Quadro drivers? WTH, nividia?

        • NoOne ButMe
        • 2 years ago

        Why are you surprised at Nvidia?

        That’s classic behavior from them. Segment, cut down, drive those professional card sales.

    • ImSpartacus
    • 2 years ago

    The TDPs are new, but the numbers in that table are not.

    [url<]https://forum.beyond3d.com/posts/1987142/[/url<] They are old stuff from AMD. Combined with that classy "Half Precision (FP32)" typo, I'm confident that the table is a custom job from that retailer.

    • flptrnkng
    • 2 years ago

    The liquid cooled performance figures are not cited. Read the fine print.

      • Jeff Kampman
      • 2 years ago

      We should. Corrected.

Pin It on Pinterest

Share This