Rumor: AMD Radeon RX 480 clock speeds leak

The rumor mill is all abuzz about AMD's upcoming Radeon RX 480 graphics card and its Polaris GPU. Until now, there was no information about the RX 480's clock speeds, even though we're mostly aware of all its other specifications. That may be about to change, though. A PC Perspective user apparently spotted a Newegg product page with a listing for an actual third-party RX 480 card, complete with detailed specs.

The product page has unsurprisingly been pulled down, but not before the reader grabbed a screenshot. The listing detailed a VisionTek RX 480 card packing 8GB of RAM. More importantly, the purported card's GPU runs at 1120 MHz base and 1266 MHz boost clocks. While barstool technicians may decry these clock speeds as low when compared to GeForce GTX 1070 and GTX 1080 cards, we think that that would be an apples-to-zombies comparison at best. In any case, the rumored clock speeds mesh well with AMD's claimed throughput of "over 5TFLOPS" for the RX 480, too.

The VisionTek RX 480 card looks quite plain as things go these days. The heatsink and fan combo is a standard dual-slot blower-type affair, and (mercifully) there isn't any RGB LED lighting in sight. We didn't see a price tag, but we'd expect this 8GB card to sell for a premium over the $200 suggested price AMD is presumably touting for the 4GB RX 480. AMD has made big claims about Polaris' power consumption, so there's a fair chance that spiffier card designs with dual and triple-fan setups could offer some interesting factory overclocks, too. We'll just have to wait and see when the wraps come off June 29.

Comments closed
    • sweatshopking
    • 4 years ago

    Only mobile matters.

      • Kretschmer
      • 4 years ago

      Why?

    • anotherengineer
    • 4 years ago

    [url<]http://www.techpowerup.com/223594/sapphire-reference-radeon-rx-480-taken-apart-pictured-some-more[/url<] Maybe the reason for the clocks is that it is not using a fancy vapor chamber designed cooler like the GTX 1080. But I guess it helps keep the costs down.

      • Spunjji
      • 4 years ago

      I’m pretty much set on getting on of these and water-cooling it for an SFF build; those photos just confirmed that it’ll fit. Good times!

    • slaimus
    • 4 years ago

    It will be interesting to see the stock situation compared to Pascal once launched.

    Since the GPUs themselves are probably in short supply, it definitely makes sense to release the higher margin 8GB cards first.

    • ImSpartacus
    • 4 years ago

    I thought this was old news. I guess I got duped by a leak that ended up being correct.

    • crabjokeman
    • 4 years ago

    No RGB LED lighting?!! That’ll shrink your epenis by two inches.

      • smilingcrow
      • 4 years ago

      Great, maybe I will now be able to get through the door of my bedroom in my parents’ basement and see if the world out there really exists.

      • CuttinHobo
      • 4 years ago

      RGB LEDs are so 2015. CMYK LEDs are the wave of the future!

    • Leader952
    • 4 years ago

    How well does TFlops correlate to actual gaming performance?

    It seems like AMD (and others) seem to be only focusing on TFlops to keep us from looking behind the curtain.

      • chuckula
      • 4 years ago

      One thing that Nvidia was accused of doing with Pascal was “brute forcing” additional performance instead of having an “efficient” architecture like GCN.

      However, in the compute side, I’m not sure that’s true. The raw compute power of a GTX-1080 is about 9 TFlops (all numbers single-precision). Certainly not a small number [b<]BUT[/b<] compared to the earlier R9 Fury-X at 8.6 TFlops, it's actually less than a 5% [i<]theoretical[/i<] advantage. However, there's a theoretical maximum that the hardware can give you when everything is running at 100% utilization vs. the real-world performance that you get in imperfect scenarios. The GTX-1080 -- even in almost ridiculous situations like the AoTs "Crazy" presets that are basically compute shader benchmarks with no useful graphical output -- still beats the Fury X by comfortably more than the 5% margin you would expect just reading the TFlop numbers.

        • AnotherReader
        • 4 years ago

        I think the discrepancy between performance on almost completely shader bound benchmarks can be explained by the GTX 1080’s high boost clock. It can boost as high as almost [url=http://www.hardocp.com/image.html?image=MTQ2MzQyNzQ1OHhlcG1yTFY2OHpfNV8xX2wuZ2lm<]1900 MHz for short bursts and can consistently exceed 1750 MHz[/url<]. Besides, I am not sure if even Ashes is completely shader bound; the Fury X only has a [url=http://www.anandtech.com/show/10326/the-nvidia-geforce-gtx-1080-preview/2<]31% increase in fps over the 290X as opposed to the 53% increase in shader throughput[/url<]. The GTX 1080 has a big edge in all other metrics.

      • Voldenuit
      • 4 years ago

      About as useful as linpack is for CPU performance.

      I.E. There is some correlation, but it is heavily workload- and case- dependent, and no substitute for detailed benchmarks.

      For shader-intensive workloads, though, it would be a decent linear correlation assuming no other bottlenecks (memory bandwidth, thread stalls, resource allocation, ROP utilization, etc).

      • ronch
      • 4 years ago

      IMHO these TFlops figures really only mean anything in the GPGPU space, a space where AMD doesn’t play in to compete with Tesla.

      For gaming, just go with the benchmarks or, in the case of upcoming products with no bench data, roughly extrapolate final performance figures from the number of shaders and clock speeds of existing products from the same vendor.

      • jts888
      • 4 years ago

      Peak T/GFLOPS in a GPGPU don’t even necessarily correlate well to peak scientific calculation rate, since the ability of a GPU and its shader compilers to keep the pipeline saturated is extremely task dependent.

      Of particular concern is code with any sort of branching, since you either have to execute all branches and mask off the untaken instructions (wasteful), or you have to dynamically regroup warp/wavefront/whatever thread clusters according to what branches each thread takes (complex and hardware intensive).

      [b<]Edit - [/b<]Note: this is only the case with so-called SIMT architectures like GCN and post-Fermi Nvidia and not for older VLIW platforms, which are poorly suited for GPGPU applications. You also have situations in games where fixed function units like ROPs, TMUs, tessellators, etc. bottleneck performance, which can be abused as is widely considered to be the case with Hair/GameWorks on non-Maxwell hardware.

      • wingless
      • 4 years ago

      You can only really compare AMD to AMD TFLOPS. Comparing them to Nvidia’s is pointless. We can compare to existing 300 series AMD cards to get a good idea of what it means to us.

      • NoOne ButMe
      • 4 years ago

      For GCN 1-3 they ended up about equal to [s<]Nvidia[/s<] Maxwell in games* when having 20% more theoretical FLOPs. With Fiji it is closer to 40%. Initially Kepler was ahead of GCN some, but GCN 1-3 is about equal or slightly ahead of Kepler today.

      • tipoo
      • 4 years ago

      Not great as a pinpointer, keep in mind it’s a mathematical number and nothing like a tested one. For AMD architectures, the equation is (stream processors) X (frequency in GHz) X 2 (Operations per cycle, 2 FMA) = GFLOPS.

      So you see, this doesn’t accommodate for ROPs, TMUs, memory bandwidth, L1 and L2 cache, memory controller differences, stream processor efficiency, or any number of the thousands of things that go into designing a GPU.

      What it does do, and I use it for often, is giving a ballpark. A 1.1Tflop GPU isn’t always faster than a 1Tflop GPU, but you know at least that the 1.1Tflop one probably isn’t going to be 5x faster than the 1Tflop one.

      Also, AMDs architectures are more floppy for the same performance level as Nvidias, for the last several years. I don’t put much thought into that, just end power draw and performance matter in the end.

      • joedacro
      • 4 years ago

      TFlops are a rating of performance. more equates to better capabilities. this doesn’t mean more fps though.

    • rudimentary_lathe
    • 4 years ago

    I’ve been following the rumours pretty closely because a) I enjoy it, and b) I’m looking for a new mid-range card.

    This is my read of the rumours right now now:
    – 5.83 teraflops at the default boost clock (2304*1266*2) – that’s R9 390X territory
    – if GCN 4.0 bring architectural improvements, the card should perform better than an R9 390X or GTX 980.
    – the GPU core is rumoured to be overclockable up to the 1.5GHz range, which assuming a corresponding linear increase in performance would put it between Fury and Fury X territory.
    – the 8GB card will cost $229USD, and AIB cards with better coolers will likely cost more than that

    I, for one, AM EXCITE 🙂 Hopefully this is how things shake out, but this wouldn’t be the first time a new GPU series got over-hyped.

      • Kretschmer
      • 4 years ago

      If it performed that well, why would AMD sell it for $200?

        • YukaKun
        • 4 years ago

        1.- Market share.
        2.- Needs the money.

        More on point 2: remember The Simpsons episode where Flanders had the lefty store?

        Cheers!

          • blahsaysblah
          • 4 years ago

          ? I read [url=https://en.wikipedia.org/wiki/When_Flanders_Failed<]wiki[/url<] page. Whats the lesson?

          • Kretschmer
          • 4 years ago

          AMD needs the money too much to give up margins. They can only make so many cards, and wouldn’t forgo “optimal profit” pricing for marketshare. That would be sacrificing survival today for long-term benefits like game developer support.

          Their low margins are generally driven by some sort of deficiency.

        • blahsaysblah
        • 4 years ago

        Only way to get back market share? $200=4GB, $230=8GB MSRP.

        • rxc6
        • 4 years ago

        Maybe because it worked well for them with the 4850/70?

        • Billstevens
        • 4 years ago

        When you are playing from behind you can’t price gouge like Nvidia.

        • Spunjji
        • 4 years ago

        It’s a fair question. It seems like there’s no reason it shouldn’t perform that well unless they fundamentally broke GCN, though, unless it’s down to a memory bandwidth limitation.

      • bfar
      • 4 years ago

      I’m enjoying this too! If this card is in Fury territory at that $230, it could even make the GTX 1070 look like a halo product. It would be interesting to see if it impacts demand for pascal cards.

    • Leader952
    • 4 years ago

    Since when did AMD get CUDA cores?

    [url<]http://www.pcper.com/image/view/70678?return=node%2F65605[/url<]

      • tipoo
      • 4 years ago

      Wow, it has 2300 CUDA cores AND 2300 Stream Processors! Can’t wait to SLI two of them on Sata Express.

        • ronch
        • 4 years ago

        Raja Koduri did say that Polaris will surprise. And so it will.

          • Srsly_Bro
          • 4 years ago

          Yup. We’ve been lied to this entire time that we have choice. We only have one GPU Overlord.

          • ImSpartacus
          • 4 years ago

          And fiji is an overclocker’s dream…

        • jihadjoe
        • 4 years ago

        2xRX 480 in RAID-0?

    • Srsly_Bro
    • 4 years ago

    I got a screenshot of that last night and it was taken down shortly after. Sapphire also kind of confirmed cards clocked at 1500 MHz would be released, too.

      • tipoo
      • 4 years ago

      That would close in on 7Tflops if it stayed there.

        • Leader952
        • 4 years ago

        <Delete>

          • Srsly_Bro
          • 4 years ago

          -Dalek?

            • Leader952
            • 4 years ago

            Cybermen

            • Srsly_Bro
            • 4 years ago

            I’ve watched every episode at least twice and I failed. forgive me.

    • tipoo
    • 4 years ago

    So, (stream processors) X (frequency in GHz) X 2 (Operations per cycle, 2 FMA) = GFLOPS, for AMD uArchs, so the boost clock gets it closer to 6Tflops, though the base is closer to 5. Around a range of 5.2-5.8.

    Now the question is how much it sticks to boost vs base.

    • EzioAs
    • 4 years ago

    [quote<] The heatsink and fan combo is a standard dual-slot blower-type affair, and ([b<][i<]mercifully[/i<][/b<]) there isn't any RGB LED lighting in sight[/quote<] I didn't know you speak French.

      • morphine
      • 4 years ago

      Je talke multiple línguas, and je sais insultos en todas ellas.

      (… for varying amounts of “talk”.)

        • Forge
        • 4 years ago

        Maintenant, je sais que peu importe à quel point mon français est, il y en a un qui est bien pire.

          • OneShotOneKill
          • 4 years ago

          Mais bon, ce n’ ai pas techreport.fr

          • chµck
          • 4 years ago

          ommlet du fromage

            • NeelyCam
            • 4 years ago

            Okei nyt ĺoppuu tollanen pelleily – kohta tulee viikingit veteleen kaikkia nassuun

            • anotherengineer
            • 4 years ago

            Ahhh Finlanders

            Don’t talk with a mouth full of Pulla please!!

Pin It on Pinterest

Share This