AMD, Breaking, Computex Taipei, Gaming, Graphics, Hardware, News

AMD Polaris 10 GPU Specifications Revealed?

A technology rumor website recently published a story detailing the upcoming AMD Polaris 10 GPU, allegedly named “67DF:C4”. As it turns out, a bird sent us screenshots showing the details of the card. While we cannot go into more details in order to protect the source, we can confirm that AMD Polaris 10 engineering samples are varying in clock between 800 and 1050 MHz, depending on the partner. Our original Polaris 10 story is here.

What we know so far is that Polaris 10 used to be known as Ellesmere, Baffin is Polaris 11 and Greenland is Vega 10. Apparently, the numbers 10, 11, 12 will not co-relate from generation to generation, but rather serve the purpose as “ArchitectureName FirstPartfromtheGPUfamily”. Polaris 10 is probably coming to the market as Radeon R9 480/490 (with or without suffix “X”), and we might see a dual GPU card as well (suffix “x2”).

AMD Polaris 10 engineering sample.

AMD Polaris 10 engineering sample.

Polaris 10 GPU “67DF:C4” Specifications

  • 14nm FinFET, GlobalFoundries
  • Diffused in USA (New York state)
  • Assembled in Taiwan
  • 2304 Cores (silicon: 2560)
  • 36 Enabled Core Clusters (silicon: 40)
  • New GPU Architecture (not GCN “1.5”)
  • 256-bit Memory Controller
  • 8GB GDDR5/GDDR5X (when available)
AMD Graphics Manager showing Polaris 10 GPU

AMD Graphics Manager showing Polaris 10 GPU

Depending on the sample, the memory is clocked at 1.25-1.50 GHz QDR (5000-6000 MHz), which the 256-bit memory bus converts to 160-192 GB/s. This however, probably won’t be the final shipping clock. Memory is estimated to reach as high as 1.75-2.0 GHz QDR stock (224-256 GB/s), which should boost the performance nicely, enabling the card to replace the old 512-bit Hawaii and Tonga GPUs. These new cards should to the market as Radeon R9 400 Series cards, even though we’re starting to hear that Radeon Pro Duo is just the beginning for the renaming of the whole product stack, which should be more in tune with the consumers.

According to sources in the know, GlobalFoundries really “kicked butt” with the switch to 14nm FinFET process, and the initial cautious approach to the new process node and transistor design turned out to be unnecessary. Today, AMD is tuning clocks above 1.05 GHz, and the final configuration might be as high as 1.15 GHz for the stock clock.

It is expected that AMD will launch its Polaris-based family of GPUs on Computex 2016 (May 31-June 4) or E3 2016 (June 14-17), in a repeat of last year’s dual launch.

  • Ace66696

    Great news ;p
    HA first

  • bedscenez .

    hope that polaris 10 r9 490x is on par with gtx 980ti..

    • γιαννης

      390 is on par with 980ti when it comes to dx12 in some cases it even surpass it..

    • Doc Holiday

      Well, it really matters how AMD prices the 490X. “If it’s just on par with a 980 ti” and they try to charge $500, it’s likely going flop against the 1080 or w/e Nvidia decides to call it, I think the 1080 will be around 20%+ faster than the 980 ti “clock to clock” with a launch price of around $500-$550. I also think the 1070 will be close to the 980 ti performance wise and go for $350. So, what I’m really hoping for is.. The 490 non X being close/ on par with a 980 ti and the 490X beating the 980 ti by at least 20%.

    • bedscenez .

      Maybe it will outperform or achieve similar performance with 980ti but with improved TDP i think..

  • Arafat Zahan Kuasha

    Finally GlobalFoundries doing something right..

    • Anonymous

      More like Samsung doing it right.

  • Pec

    So will it outperform the Fury X or not? Does HBM still have the advantage for 4K and 3440×1440 users?

    • Rauel Crespo

      Well, lets say their’s a Big Polaris 10 (R9-480x) and a cut down Polaris (R9-480), seeing as there are two entries for both cores in clusters in the specs means this is realistic.

      So, I would venture to guess that the R9-480-4GB will be $250 and offer about R9-390 performance with 4GB of GDDR5. That’ll probably be the first card out.

      Then when GDDR5x is available in quantity, AMD can release an R9-480-8GB, with the same Chip, but closer to R9-390x/Fury performance for $330.

      And finally would be the R9-480x, a full Polaris 10 with 8GB of GDDR5x, for $410. Performance will be between Fury and Fury-X.

      • effin

        That sounds extremely disappointing. Coming up with a new generation of GPU’s on the biggest node shrink (from one gen to the next) we’ve ever had and you can’t even provide ONE card to be above the top one in last gen?

        Nvidia easily did this on the SAME 28nm node with the 900 series, where even the 970 is trading blows with the 780ti. The 480x (full polaris 10) should be head and shoulders above FuryX or they will be absolutely smashed by nVidia… AGAIN!

        • Ashaira

          that is false. for a die shrink/new architecture the power of new cards is shifted around 2-3 tiers not more. a 480x being as fast or slightly faster than a fury x while costing half or less is more than a decent enough. keep in mind that new cards don’t just bring better performance but new tech as well.

          • Conall Ó Gribín

            The overclocking headroom that comes with speed binned first gen parts can be crazy too. Maybe not as crazy as their CPU counterparts (how I loved the E6300 I bought and overclocked by 55% on air when conroe was just launched).

            People need to remember that its in their best interest to be conservative with guaranteed operating speeds for new parts based on new architecture!

        • Anonymous

          Actually, Nvidia will do the same. Not releasing the big die for consumer yet, because the big die will be HBM2 and it’s too expensive right now.

          By the time Nvidia releases GP100, AMD will bring Vega, which will be a massive jump in performance.

          Oh, and you should check some reviews, as the Fury X beats Titan X and 980ti in 7 out of 8 DX12 games. So AMD clearly has the GPU performance crown at the moment.

          • effin

            “Actually, Nvidia will do the same. Not releasing the big die for consumer yet”

            But this is EXACTLY what nVidia did with the 900 series as well! They didn’t release the big die for consumer at first, only the 970 and 980 and STILL managed to outperform the big die from the 700 series. And this on the SAME node!!! AMD not having one single card above the Fury-X at launch will be a COLOSSAL fuck up and nVidia will eat them alive, there will be no coming back from this fail.

            Luckily, if AMD didn’t lie in that 1440p 60fps Hitman benchmark (if high settings) it looks like they will have at least a card to outperform the Fury-X.

          • Anonymous

            28nm node was mature at that point, so Nvidia designed their GM204 relatively big too. Just a third smaller than GM200 and only 29% smaller than GK110.

            16nm will be a bit more difficult to get good yields in the beginning. It’s expected that their second biggest die will be considerably smaller, to be able to sell it as a gaming GPU at an acceptable price point.

            However, I do expect Pascal (second biggest die) to outperform Polaris (bigger of the two dies), slightly. The focus is more on power efficiency, new features, etc. for now.

            The real performance showdown will be big Pascal vs Vega, which are both big dies with HBM2.

        • BestJinjo

          What are you going on about? Polaris 10 is NOT AMD’s highest end chip/card of this generation. Did you not see AMD’s road-map?

          There will be Vega 10 and 11 chips with HBM2.
          http://cdn.videocardz.com/1/2016/03/AMD-Radeon-Polaris-Vega-Navi-GPU-Roadmap-900×506.jpg

          http://videocardz.com/58564/amd-announces-gpu-roadmap-for-2016-2019-polaris-to-be-followed-by-vega-and-navi

          Chances are Polaris 10 will not cost $649 which is what true next gen flagships from AMD/NV will cost (i.e., Vega 10 and Big Pascal GP100/102).

          Both AMD and NV are launching mid-range chips first, exactly the same strategy used in the 680/7970 generation. Then bigger chips will come out later in 2017. This is exactly like AMD/NV releasing 290X/780Ti.

          • effin

            Well, definitely you don’t know what i’m on about. I’m not saying polaris 10 is the highest end of this generation. I don’t even know what made you understand such a thing. I’m saying that at least one Polaris 10 card should beat the Fury-X, as was the case of 970/980 vs 780ti, which were mid range cards. And that happened on the same node. Now it would be extremely bad for AMD if not even one of their cards will be above 980ti.

          • While the exact Polaris 10 / 11 details are not available, IMHO AMD might release a dual-Polaris 10 card as their 2016 high-end part. Polaris 10 is probably hitting the $399-449 price bracket, given that 490/490X is supposed to succeed the 390/390X family of products. If they were targeting Fury, the product would be a part of Fury product family.

        • Dan

          Fury x and 980ti cards make up very little of the GPU base. Cards selling for 150 to 300 make up the majority of all cards sold to PC gamers. So it is most important that those cards perform well. Not that they have a card that is better than the 980ti out of the door.

          That being said . . . It really depends on the yeilds you can get on dies as far as how large they can go. Video cards are largely about how many stream processors you can pack onto a die, and then how large you can make that die.

          The card in this article has 2560 stream processors which is the same as the R9 390. So assuming they were able to make the processors more efficient, the card shown here will perform slightly better than a R9 390. But the die size on the 390 was 438mm2. so this die would be roughly ~215mm2. Will also have significant efficiency improvements. Who knows with this card is though . . . it could be the R9 470 (the 370 has a die size of 212mm2) . . . No one really knows the yeilds on 14nm finfet and how large you can realistically make dies.

          As long as the technology is somewhat decent though, you can just disable cores that have defects. and sell those as 470s and sell pristine dies with no defects as X editions . . .

          Polaris 10 should be able to beat fury though. they will need to make a die roughly 300mm2 in size . . . which doesn’t seem too far fetched.

          you really cannot know until launch though . . . and you can get some unbiased benchmarks . . . until then it is a pure guess. This entire article might be 100% false . . .

    • Vega is the Fury replacement. Polaris is the mainstream stuff.

  • TheDizz

    @theovalich:disqus You adding in your own speculation about the naming scheme 480/470 or is that based on info from your sources? Just clarifying because that wasn’t really clear from your article.

    • Based from the sources I talked to.

      • TheDizz

        Interesting. Thank you.