AMD Radeon Pro Duo bridges the professional-consumer divide

AMD first showed off its Radeon Pro Duo graphics card back at GDC, but the company’s Capsaicin event left a number of unanswered questions about the newest in AMD’s long line of single-card, dual-GPU solutions. The Radeon Pro Duo officially launches today, and we now know all of its secrets. One thing we unfortunately won’t be discussing is detailed performance results. As you may have already read elsewhere, AMD didn’t send Pro Duos to many of the usual suspects in the PC hardware press for review, ourselves included.

Some of our readers may want to put down their pitchforks after reading that last sentence. You see, AMD isn’t positioning the Pro Duo as an enthusiast graphics card in the vein of the Radeon R9 295 X2. Instead, the company tells me the Pro Duo is intended as a multi-GPU development platform for VR content creators and other content professionals—people who game, to be certain, but whose time at the keyboard is mostly spent building those games (or using other professional applications that take advantage of GPU acceleration). In that sense, the Pro Duo is trying to be a bridge between the workstation- and consumer-graphics worlds, rather than the baddest single-card graphics solution around for gamers and enthusiasts.

That’s not to say the Radeon Pro Duo isn’t an exciting product—it is—but this card’s $1499 price tag and professional bearings mean it probably won’t find its way into any but the most dedicated of enthusiasts’ desktops. Even if only a lucky few will be pushing pixels with a Pro Duo, it’s worth touching on some of the achievements AMD has made with this product.

Double the Fiji, double the fun

First off, this card houses two fully-enabled Fiji GPUs, as seen on the Radeon R9 Fury X. Unsurprisingly, its spec sheet reads a lot like two of those cards smashed together: 8GB of HBM RAM, 8192 stream processors, 128 GCN compute units, and dizzying potential memory bandwidth of up to 1024GB/s, just to mention a few highlights. As we’ll see in a moment, all those resources mean the Pro Duo has prodigious potential performance available.

One difference from the Fury X is that those Fiji chips only run at “up to 1000 MHz.” That’s 50 MHz down from the Fury X’s peak boost speeds, and similar to the way AMD describes the performance of the fully-endowed Fiji chip in the Radeon R9 Nano. As a result, a pair of those diminutive cards in CrossFire might serve as a rough yardstick for the Pro Duo’s performance. The card’s thermal envelope of 350W is exactly twice that of a Radeon R9 Nano, too. While the Pro Duo will need three eight-pin power plugs to satisfy its thirst for energy, its board power specification is still well short of the R9 295 X2’s monster 500W figure.

As it did with the R9 295 X2, AMD has once again turned to liquid cooling to keep the heat of those twin GPUs in check. We’d love to tear into some of the plastic shrouds covering the pumps and water blocks on this card, but all we can do is gaze from afar for now. Even in pictures, however, the Pro Duo’s cooling system looks more sophisticated than the one on the Fury X. Going by appearances, I’m guessing the cooling system uses something like full-coverage blocks over the power circuitry, versus the simple copper coolant pipe running over the VRMs of the older card. Once the hoses leave the case of the graphics card, they run to a 120-mm radiator similar to that of the Fury X’s.

The rest of the Pro Duo looks quite a bit like a lengthened Fury X. This card maintains the same fancy metallic frame, rubberized covers, and LED illumination as its sibling. We’re betting AMD’s board partners will have the same minimal room for brand-specific modifications as they did on the R9 Fury X. Expect manufacturer stickers on the fan hub and little else in the way of manufacturer customizations. AMD retained the “modular faceplate” of the Fury X on the Pro Duo, though, so crafty owners will still have some room for personal expression.

One minor change from the prototype shown at AMD’s Capsaicin event is the removal of one of the card’s four DisplayPort 1.2 outputs. That port has been replaced with an HDMI connector, presumably to improve compatibility with VR headsets like Oculus’ Rift that require an HDMI port to plug into a host PC.

 

Two chips are better than one (sometimes)

The big number AMD likes to throw around when talking about the Pro Duo is its 16 TFLOPS of theoretical single-precision (or FP32) compute performance. Nvidia doesn’t have a comparable dual-GPU product on the market right now, so the fairest competitor here may be the green team’s fastest single-card consumer product, the GeForce GTX Titan X. That card can deliver theoretical single-precision compute performance of 7 TFLOPS. For a different perspective, the Pascal-powered Tesla P100 can crunch FP32 data at 10.6 TFLOPS, but that’s not a product any of us will be putting in our desktops any time soon.

AMD suggests a couple scenarios for putting all that raw performance to use. For one example, the company says applications using its FireRender SDK can get a big boost running on a single Pro Duo. Going by AMD’s numbers, rendering a single example frame in Autodesk’s 3DS Max on the Pro Duo takes 15 seconds on the fully-enabled  dual-GPU card, compared to 26 seconds with one of the Fiji GPUs disabled and 228 seconds on the host system’s Core i7-5960X CPU.

In another scenario (this time without FireRender), the company says Blackmagic’s Davinci Resolve 12 color-grading suite could run twice as many nodes of radial blur on a Pro Duo-equipped PC while maintaining smooth preview playback when compared to a single Fiji GPU running the same task. Presumably, that means colorists or other video pros can slather on the effects during editing without sacrificing smoothness.

All that’s well and good, but AMD calls the Pro Duo a card “for gamers who create and creators who game.” The company isn’t shy about calling the Pro Duo “the world’s fastest graphics card,” and that label may be justified. The company released some average-FPS benchmarks from a variety of titles running at 4K to justify this claim. By that measure, the card certainly puts some distance between itself and the Radeon R9 295 X2, not to mention the single-GPU Titan X. The graphics settings the company chose to demonstrate those results were pretty typical of what we’d choose ourselves, too, not the weirdly scaled-back ones we’ve sometimes seen in AMD’s internal testing.

We’d exercise caution when drawing any major conclusions from these numbers one way or another, though, thanks to the inherent coarseness of the average frame rate measurement.  We’d also like to have seen the card pitted against two Titan Xes in SLI, since AMD wasn’t shy about using that exact setup for its Ashes of the Singularity testing. Still, the numbers AMD shared hint at the Pro Duo’s considerable performance potential. Whether games take advantage of that potential is another story—many modern titles don’t support alternate-frame rendering—so there may be a bit of a question mark hanging over the Pro Duo’s value for the traditional 3D titles of the future.

The Pro Duo’s real forte, then, may be its performance in VR. The card is the only product that’s in AMD’s “Radeon VR-Ready Creator” tier so far, and it seems a natural fit for use with the company’s Affinity Multi-GPU feature (even if that technology isn’t widely supported yet). For now, though, the company shared some numbers generated by the SteamVR Performance Test. The Pro Duo maxes out this test with a score of 11, followed closely by a single GeForce GTX 980 Ti with a score of 10.5. We won’t dwell too much on that result, other than to say that the Pro Duo appears to live up to its billing as a capable platform for VR content production and playback.

In keeping with the “creators who game” theme, the Pro Duo comes backed up with an interesting driver support model. Content pros can install FirePro drivers for the Pro Duo, with all of the stability and vendor-certified credentials that software offers. Pro Duo owners who game can install the same Radeon Software drivers from consumer cards to get that software’s game optimizations and cutting-edge support. It’s likely one will need to uninstall and reinstall each driver package as needed to make this dual-driver support model work. We’ve asked AMD for more information regarding this setup.

AMD says the Radeon Pro Duo will be available from “select” add-in board partners and system integrators today for $1499. Depending on what Radeons you consider comparable to the Pro Duo, acquiring a pair of Radeon R9 Nano cards for use in CrossFire costs about $500 less, while a pair of Fury X cards narrows that gap to $250. For the extra money, the Pro Duo gets buyers two Fiji GPUs on one card, requires only one radiator mount for cooling purposes when compared to Fury Xes in CrossFire, and offers a distinctive value proposition: access to AMD’s workstation and consumer graphics drivers alike. For folks who often need to move between the worlds of content creation and content consumption, then, the Pro Duo might be just the ticket. The rest of us might be better-served by keeping an eye on Polaris.

Comments closed
    • Chrispy_
    • 4 years ago

    [quote<]creators who game[/quote<] I know several people in the game industry and those of them who can be classed as creators could never afford this card themselves. They barely get paid an average wage, let alone a good one. Not just the self-employed and indie developers, but also the content creators working under large developers would benefit from having affordable access to FirePro and Quadro drivers on their affordable Radeons and Geforces. I know it would hurt sales of their professional cards, but with all the major CAD vendors opting to choose broad compatibility so that they can run on the IGP of ultrabooks and convertibles, their high-end, high-price "professional" cards are already being shunned by the professionals en-masse. Time to look at the big picture and remove the artificial product divide maybe?

    • tsk
    • 4 years ago

    Review out over at pcper.com
    TL;DR Pretty much what was expected, R9 nano x2 and Nvidia still has better frame times.

    • Mr Bill
    • 4 years ago

    At least, this version is cheaper than the $6000 [url=http://www.anandtech.com/show/10209/amd-announces-firepro-s9300-x2<]AMD FirePro™ S9300 x2 Server GPU[/url<] see also [url=https://techreport.com/news/29926/amd-firepro-s9300-x2-is-a-radeon-pro-duo-for-the-hpc-set<]TR story[/url<]. I guess this reflects the difference between rated for actual workstation use versus gaming development.

      • Mr Bill
      • 4 years ago

      Doh! chuckula already said this earlier in the thread.

    • south side sammy
    • 4 years ago

    all I want to see at this point is what operating systems support this card.

    • psuedonymous
    • 4 years ago

    “The Pro Duo’s real forte, then, may be its performance in VR.”

    VR multi-GPU does not scale magically due to latency constraints: you can’t gain scaling efficiency by having multiple frames in flight at one time, you need both GPUs to work on the same frame. From Alex Vlachos’ (Valve) Advanced VR Rendering talk at GDC 2016: [url<]http://i.imgur.com/iVXQo5t.jpg[/url<]

      • AnotherReader
      • 4 years ago

      Dedicating one GPU to each eye, ala Affinity Multi-GPU, will make good use of the Radeon Pro Duo. Of course, it requires work; even normal crossfire/SLI requires work.

        • psuedonymous
        • 4 years ago

        It will make less use of it than AFR would. See the slide, scaling is limited to parallelisable tasks, which at the moment give a a 30-35% performance increase for adding an additional GPU (and a MUCH smaller increase for adding another two above that). That’s not a great increase in performance for a more than doubling of cost (the Pro Duo is priced above 2x Fury Xs).

          • Waco
          • 4 years ago

          He’s talking about having 100% performance for each eye via dedicated GPUs. Why do you think that won’t scale nicely?

          • AnotherReader
          • 4 years ago

          I looked at the slide and it is informative. However, traditional AFR doesn’t yield 100% increases either. It seems that much of the work is sped up by 100%. Transferring the frame to the primary GPU and application overhead account for the slowdown. The transfer latency is also part of AFR. I haven’t seen the talk, but do you mean a 30 to 35% decrease in frame latency? If that is the case, then that is a 40% to 50% performance increase.

    • HisDivineOrder
    • 4 years ago

    AMD is AMD. What do you expect? AMD loves to selectively decide products don’t “belong” to reviewers or not. They’ve been doing it with CPU upgrades for years. GPU’s are just getting started…

    If you can’t bully the reviewers into more positive reviews, you have to prevent the reviewers from getting them at all.

    Naturally, Twitch and Youtube streamers will wind up with boatloads of them.

    • Klimax
    • 4 years ago

    How many (VR) devs will buy this without reviews? I doubt it will get anywhere even in supposedly intended market.

    • Umbral
    • 4 years ago

    I’m not getting a good vibe from AMD – this year’s products look either down-market or (like this card) just plain baffling. At this point I’m hoping Nvidia doesn’t feel like milking me for every penny when Pascal Jr. hits the streets.

      • muxr
      • 4 years ago

      We’ve reached the maximum manufacturability when it comes to packing Tflops into a GPU on 28nm node. And the yields are increasingly lower as we shrink further.

      When you consider that VR requires maximum performance crammed into a GPU, and that 2 GPUs make absolute sense since you are powering two different “monitors”, it becomes crystal clear who this card is for.

      It’s for VR content creators, “VR experience” creators (think theme parks), developers, crypto miners, OpenCL high speed hashing.. there are tons of specialized uses for this card.

      Now consider, as much as no one wants to hear it, that for the gaming hardware to be sustainable it has no choice but to go to multi GPU. Multi GPU is the future, and currently we’re in a sad state of affairs, with both xfire and SLI, the DX12 and Vulkan Linked Multiadapter is the future.

      This is why this card exists. For those developers to work on content we will see as the Multi GPU gets wider adoption.

    • anotherengineer
    • 4 years ago

    Meh

    My vid card budget lies in the $175-$250 range.

    It would probably kind of perform like this.
    [url<]http://www.tweaktown.com/articles/7226/amd-radeon-r9-fury-video-cards-crossfire/index5.html[/url<]

    • WaltC
    • 4 years ago

    Next gen (14nm FinFet) is where the real fireworks will begin, imo, along with HBM2 implementation. This might be a nice buy for developers providing it’s the exact same thing as developing for xFire–which it should be, of course. Be nice to see some game engines come out that support d3d12’s more interesting features, like being able to use this 8192 mb pool of ram completely, instead of as using it for twin 4096 mirrored pools, etc., as must be done with earlier API support.

    • Chrispy_
    • 4 years ago

    I’m pleased that AMD is targetting the 175W per SKU. It’s a figure I’ve heard rumoured for Polaris too.

    What this means is that quiet dual-slot cooling should be easy and it should be easy to find reference models with sensible power, heat and noise levels.

    Way back last year I posted my experiments with underclocking and underpowering my Hawaii cards, and it proved to me that AMD boosted the hell out of their 290/390 line (also non-Nano Fury cards) by driving them at the raggedy edge of their voltage and clock limits.

    Turns out you can get massive power/heat/noise savings by just reducing the clocks a little bit – My 1050MHz Hawaii still ran at 925MHz using something like 100W less power.

    • Tristan
    • 4 years ago

    Give us real tests, not slideware propaganda from AMD.

      • End User
      • 4 years ago

      Calm Down:

      “One thing we unfortunately won’t be discussing is detailed performance results. As you may have already read elsewhere, AMD didn’t send Pro Duos to many of the usual suspects in the PC hardware press for review, ourselves included.”

      • beck2448
      • 4 years ago

      No cards for independent reviewers???? That doesn’t sound good.
      Watercooled Crossfire. Pass.

    • Krogoth
    • 4 years ago

    Methinks this is mostly “vaporware”.

    It is AMD’s attempt to have the #1 gaming product on the market until Pascal comes along. There’s nothing professional about this product. If were to be case then it would have been marketed under the “FirePro” family. At best, it is AMD’s equivalent to Nvidia’s “Titan” line-up.

      • muxr
      • 4 years ago

      It’s a very important card for developers. As AMD is going the multi-GPU route for consoles and VR, this card was needed in order to provide the developers and content creators with the early tech they can develop on.

        • chuckula
        • 4 years ago

        [quote<]It's a very important card for developers. [/quote<] Proof? [quote<]As AMD is going the multi-GPU route for consoles and VR[/quote<] Horse pucky. You are literally schilling in favor of today's position that these multi-GPU cards aren't for consumers while in the same breath claiming that cheap consoles are suddenly going to be using multiple piece of GPU silicon? Then again, being an AMD fanboy these days requires a certain level of self-contradiction salted with disbelief in AMD's official positions.

          • muxr
          • 4 years ago

          Where you also a luddite about dual core CPUs when AMD introduced the Athlon 64 x2?

      • Mr Bill
      • 4 years ago

      (1) Would a dual Titan card be any cheaper?
      (2) Its not “vaporware” if you can buy it.
      (3) Its a pretty smart ‘attempt to have the #1 gaming product on the market’

      Kudos to AMD for finally making the effort to fill the top performance slot.

      • beck2448
      • 3 years ago

      One problem with the Radeon Pro Duo is it’s really not very good.
      [url<]http://www.pcper.com/reviews/Graphics-Cards/AMD-Radeon-Pro-Duo-Review/Grand-Theft-Auto-V[/url<]

      • Mr Bill
      • 3 years ago

      Now that the other shoe had dropped and the [url=https://techreport.com/news/30422/nvidia-unveils-a-pascal-powered-titan-x-with-11-tflops-on-tap<]11 TFLOPS Pascal-powered Titan X[/url<] has been revealed, it would be very cool to see these squared off with gaming and workstation loads and the newest drivers.

    • maxxcool
    • 4 years ago

    “professional”, 1500$, and not Dbl precision ? … /facepalm/

      • NTMBK
      • 4 years ago

      I give you the Quadro M6000, $5000 pro card with 5x less double precision performance than this card. [url<]http://www.newegg.com/Product/Product.aspx?Item=N82E16814133623&cm_re=quadro_m6000-_-14-133-623-_-Product[/url<] There's more to "pro" than DP prowess.

        • maxxcool
        • 4 years ago

        People BUY these things !? .. ok nm… carry on AMD… carry on.

        • Krogoth
        • 4 years ago

        It doesn’t have 10bit support or ECC memory on it.

        Radeon Pro brand = AMD’s answer to “Titan”.

        Ultra-high end boutique products carter towards who want maximum performance and budget is no object. The earlier Titans were more interesting as they were just either “failed” chips or excessive Quadro/Telsa stock that kept DP performance intact.

          • auxy
          • 4 years ago

          There’s no such thing as ECC GDDR5 memory. ECC on Quadros is done through parity emulation, which is why there’s less video memory available in ECC mode.

          All Radeons support 10-bit mode, unlike Geforces. TMYK (*´ω`*)

            • Krogoth
            • 4 years ago

            [quote<]All Radeons support 10-bit mode, unlike Geforce. [/quote<] 10-bit support only exists on FirePro line-up. The regular Radeon units do not have it unless you do a firmware hack just like the normal Geforces, but good luck getting 10-bit content to work correctly with normal Radeon/Geforce drivers. [quote<]There's no such thing as ECC GDDR5 memory. ECC on Quadros is done through parity emulation, which is why there's less video memory available in ECC mode. [/quote<] You really do not understand how ECC works. It always has take-up a portion of the your memory capacity for the "parity" table. ECC DIMMS have another chip on them for this and need tracing for an extra 16-bit bus for this. That's why ECC DIMM support is limited on a certain motherboards and controllers. It is no different on GPUs that have "ECC memory support". That's why there are no firmware hacks for for it.

      • Ninjitsu
      • 4 years ago

      It sounds more “prosumer” tbh. Also, i highly doubt they’ve turned DP off or something, it should have around 1TFLOP of FP64 (1/16th of FP32 for Fiji).

      Less than GP100 iirc, and doesn’t appear to be relevant to their target market, I suppose, would be why they (AMD) haven’t mentioned it.

        • Mr Bill
        • 4 years ago

        Yep. FP64 is 1/16 of FP/32 compute for Fiji core, which is same as for the Tonga core that is now in the FirePro lineup.

    • chuckula
    • 4 years ago

    For the fan squad here justifying everything AMD does because AMD, please have a read as to what your corporate masters said a few months ago while you twist yourselves into pretzels pretending that these cards somehow have nothing to do with playing games:

    [quote<]Q. On the E3 Livecast, Lisa committed to shipping Fiji Gemini by Xmas. What happened? Is Fiji Gemini delayed? A. The product schedule for Fiji Gemini had initially been aligned with consumer HMD availability, which had been scheduled for Q415 back in June. Due to some delays in overall VR ecosystem readiness, HMDs are now expected to be available to consumers by early Q216. To ensure the optimal VR experience, we’re adjusting the Fiji Gemini launch schedule to better align with the market. Working samples of Fiji Gemini have shipped to a variety of B2B customers in Q415, and initial customer reaction has been very positive.[/quote<] [url<]http://techfrag.com/2015/12/23/amd-confirms-r9-fury-x2-delayed-till-2016/[/url<]

      • jackbomb
      • 4 years ago

      Hey! You justify everything Intel does because Intel.

      • muxr
      • 4 years ago

      I read your post 3 times and I am still not sure what you’re trying to tell us.

        • chuckula
        • 4 years ago

        Your inability — or outright refusal — to understand the simple words of your corporate overlords is not my problem.

          • muxr
          • 4 years ago

          I understand what Lisa said perfectly fine, it’s your contradicting conclusion that makes zero sense.

          – She said the VR ecosystem wasn’t ready back then… True.

          The card is intended for VR content creators, developers and lastly enthusiasts (in that order).

          > pretending that these cards somehow have nothing to do with playing games

          Your understanding of what she said is that this card is intended for Gamers? Wut? She said no such thing.

          You make zero sense.

            • cegras
            • 4 years ago

            He’s shadow boxing the AMD fanboy as usual. I wish chuckula could be muted in the comments section. I can feel his angry rebuttal already.

            It’s amusing to watch his tone change as he hops from AMD (“For the fan squad here justifying…”) to intel (damage control, see [url<]https://techreport.com/news/30011/intel-cuts-its-workforce-by-11-as-it-releases-q1-financial-results)[/url<] news threads.

            • muxr
            • 4 years ago

            I really don’t understand what would compel someone to spend so much time dogging on the only other company in this space we all feel passionate about.

            As I type this on my Intel powered computer, I would want nothing more than for AMD to provide some good competition, to both Intel and Nvidia.

            Especially when we consider that historically speaking, a lot of the innovation came from that very company.

    • NTMBK
    • 4 years ago

    This isn’t for us (or at least, most of us). This is for if you need 2017’s performance today- e.g. you’re developing a VR game meant to be released in a year or two’s time, and want something close to your target platform.

    The rest of us? Wait for Pascal and Vega.

      • muxr
      • 4 years ago

      Spot on. Not sure why you’re getting downvoted, but developers and content creators who are working on next gen consoles and soon to be released VR products is exactly the target audience for this card.

    • chuckula
    • 4 years ago

    [quote<]As you may have already read elsewhere, AMD didn't send Pro Duos to many of the usual suspects in the PC hardware press for review, ourselves included.[/quote<] The f#%k AMD? Really? You hired Wasson and you still can't be bothered to send a review unit out to TR? I was starting to be more positive about the GPU-side of AMD given that they appeared to be making some smarter moves, but then they go and pull a [url=http://www.hardocp.com/article/2015/09/09/amd_roy_taylor_nano_press/2<]Taylor[/url<] and show that they still don't really want to win.

      • NTMBK
      • 4 years ago

      Same reason that NVidia never sent out a Titan-Z review sample- this isn’t a card aimed at TR’s audience.

        • chuckula
        • 4 years ago

        Don’t let the “pro” marketing name fool you. This is a single precision gaming card, not a double precision workstation card. I never complained that TR didn’t receive a sample of that $6000 Firepro card since that really isn’t for most of TR’s reader base.

        Remember that AMD’s own marketing department bent over backwards to claim that the massive delays preceding the release of these cards were because VR wasn’t ready yet… So here we are with AMD’s supposedly ultimate VR card that they won’t send out to TR for review!

        Oh and for the record, Nvidia was dumb not to send out the Titan-Z for review either even though they did have the legitimate excuse that it is more of a CUDA card than a gaming card. Even though the several years old Titan-Z has more than triple the double-precision power of this Pro Duo, it would have been interesting to see TR review it.

        [Edited for spelling, Android keyboard bad]

          • NTMBK
          • 4 years ago

          There are plenty of “pro” workloads that don’t require lots of double precision. Machine learning for one.

          Again, look at NVidia’s product positioning. They pushed the Titan X as a “pro compute” card, despite it having 8x less DP FLOPs than the Titan Black it was replacing.

            • chuckula
            • 4 years ago

            1. “Pro” or not [and it really isn’t], Nvidia didn’t seem to have a problem sending the Titan X over to TR for review: [url<]https://techreport.com/review/27969/nvidia-geforce-gtx-titan-x-graphics-card-reviewed[/url<] 2. If the this Pro Duo whatever is really such a "pro" card then please answer this: Why doesn't it have a standard HSF but instead requires this closed-loop liquid cooler that's clearly not designed to work with a standard workstation case? Oh, and while we are at it, how many "professionals" do you know who would give a rats 4ss about getting a non-functional souvenir Fiji die along with their supposedly "professional" card. Because that's what AMD is packaging here. Or we can step back and look at the situation in the context of reality instead of pro-AMD fanboyism and realized they screwed the pooch here.

            • NTMBK
            • 4 years ago

            I’m not trying to say it’s an ideal workstation card. It’s obviously a weird niche card with very limited appeal. Just saying that lack of DP does not disqualify something from being pro.

            And while you may quibble over whether Titan X counts as “Pro”, the Quadro M6000 certainly does. And that has the same limited DP capabilities that the Titan X has.

          • Wild Thing
          • 4 years ago

          “Oh and for the record”…..If NV had been dumb enough to have the TitanZ tested it would have been smashed by R9295X2 (at half the price)
          Pro Duo however destroys every card in Nvidia’s lineup.

          *Please continue to let the butthurt flow through you*

            • NTMBK
            • 4 years ago

            “Destroys” is a little too strong. There are tons of games where Crossfire/SLI just doesn’t work properly, in which case you might as well have bought a Nano.

      • gbcrush
      • 4 years ago

      [quote<]The f#%k AMD? Really? You hired Wasson and you still can't be bothered to send a review unit out to TR?[/quote<] Beat me to it. We want a refund. Give us back our Damage! [quote<]Some of our readers may want to put down their pitchforks after reading that last sentence. [/quote<] Nope. I've got a pitchfork...with a flaming heap of pitched covered hay on the end. Not putting it down any time soon! *starts a riot*

      • maxxcool
      • 4 years ago

      ^ +crapload … wow … Makes me think they don’t have enough stock to actually go around.

      edit: thing to think .. damn coffee jitters

    • derFunkenstein
    • 4 years ago

    [quote<] We'd exercise caution when drawing any major conclusions from these numbers one way or another, though, thanks to the inherent coarseness of the average frame rate measurement. [/quote<] AMD's penchant for weird 0xAA and 0xAF numbers are another reason to be wary. The company's benchmarks tend to differ somewhat from reality, where gamers and reviewers alike crank up the filtering and anti-aliasing options to get an image that's pleasant to the eye.

      • Jeff Kampman
      • 4 years ago

      Here’s a data dump of the settings used this time around:

      [quote<]Page 14: 4K Gaming Performance: Testing conducted by AMD Performance Labs as of March 7, 2016 on the AMD Radeon™ Pro Duo, AMD Radeon™ R9 295X2 and Nvidia’s Titan X, on a test system comprising Intel i7 5960X CPU, 16GB memory, Nvidia driver 364.72, AMD driver 16.4.1 and Windows 10 using Rise of the Tomb Raider v1.0_638.8_64 DX11 Foundation benchmark 3840x2160 veryhigh FXAA 16xAF quality=veryhigh api=DX11; GTA V 335.1 DX11 RAGE benchmark 3840x2160 ultra Preset 0xMSAA 16xAF Tessellation=veryhigh ShadowQuality=veryhigh ReflectQuality=ultra SSAO=high TextureQuality=veryhigh ParticleQuality=veryhigh WaterQuality=veryhigh GrassQuality=ultra ShaderQuality=veryhigh SoftShadows=softest PostFX=ultra DoF=true FXAA=true Quality=ultra; Battlefield 4 1.1 DX11 Frostbite 3 dam 3840x2160 Ultra Preset 4xMSAA + FXAA + off 16xAF quality=ultra mantle=off; Assassins Creed Syndicate 1.31 DX11 AnvilNext benchmark 3840x2160 veryhigh 2xMSAA + FXAA 0xAF quality=veryhigh; Ashes of the Singularity 1.00.18769 DX12 Nitrous engine benchmark 3840x2160 crazy 4xMSAA 0xAF API=DX12 Quality=crazy AsyncCompute=on;Far Cry Primal 1.2.0 DX11 Dunia Engine 2 benchmark 3840x2160 ultra preset SMAA N/A quality=ultra, to simulate GPU performance. AMD Radeon™ Pro Duo scored 46.6fps, 83.3 fps, 81.6 fps, 42 fps, 52.8 fps, 59 fps, respectively; AMD Radeon™ R9 295X2 scored 37.3 fps, 65.2 fps, 63.972 fps, 35.2 fps, 42.6 fps,52 fps, respectively;Titan X scored 37.8 fps,50.5 fps, 49.7 fps, 28.3fps,26.7 fps, 35 fps, respectively; PC Manufacturers may vary configurations, yielding different results.[/quote<]

        • derFunkenstein
        • 4 years ago

        That’s excellent, way different than the Fury X pre-release benchmark settings. I was thinking of when [url=https://techreport.com/news/28501/here-a-first-look-at-the-radeon-r9-fury-x-performance<]marketing numbers[/url<] didn't line up with [url=https://techreport.com/review/28513/amd-radeon-r9-fury-x-graphics-card-reviewed<]actual settings[/url<] you'd want to use. Glad to see a shift.

          • Ninjitsu
          • 4 years ago

          Of all the things that Scott is jokingly assigned responsibility for improving, this is indeed likely something he could have influenced.

            • derFunkenstein
            • 4 years ago

            Agreed, yet no “thanks, Scott” posts. They stop being amusing and finally one could be warranted.

            • Mr Bill
            • 4 years ago

            +++

      • brucethemoose
      • 4 years ago

      To be fair, post-processing AA has pretty much replaced MSAA.

      There’s no excuse for turning AF off though, even if it doesn’t have a huge performance hit.

Pin It on Pinterest

Share This