PC Perspective pokes and prods the Radeon Pro Duo

Last year or so, AMD showed off a prototype of an upcoming dual-Fiji card. The two-headed cat eventually came out of the bag, and it was shown to the world as the Radeon Pro Duo. While AMD initially hinted at the Pro Duo being a new breed of gaming monster, the company is now billing the card as a simple, easy-to-install solution for VR developers and other professional users. There's a burning question, though: "how well does it game?" Well, PC Perspective got a hold of one of these babies and put it through its paces.

The Radeon Pro Duo has two full-sized Fiji GPUs and 8GB of HBM RAM. That sounds oddly close to a couple of R9 Nanos, and if you think "Nanos-on-a-stick," you're close enough for most purposes. Since the Pro Duo is a high-powered card, PC Perspective ran its testing at 2560×1440 and 3840×2160 with all the pretties turned up. The main comparison points we'll be looking at are a pair of Radeon R9 Nanos and a similar set of GeForce GTX 980 Ti cards. Let's see how things shook out.

The Radeon Pro Duo showed off its grunt in Fallout 4. Average FPS was quite high even while playing at 4K, often staying around or above 60 FPS. As we repeatedly drill into everyone's heads, though, average FPS numbers by themselves are meaningless if those frames aren't being delivered steadily—or, in other words, if the game you're playing stutters or doesn't otherwise feel completely smooth. Similar-looking FPS averages can turn out to be very different stories in practice.

When it comes to frame times in Fallout 4 at 2560×1440, PC Perspective found that the Radeon Pro Duo and R9 Nano put in a respectable showing, with relatively small frame variation. The set of 980 Ti cards was better overall, but the difference looks to be nothing to write home about. The story is roughly the same at 4K resolution.

The PC version of GTA V is a very pretty game, and great graphics beget great horsepower. The Radeon Pro Duo ought to be at home here, and the initial FPS average measurements seem to indicate it is. That feeling probably only lasted until PC Perspective actually played the game, though. Put simply, the frame time variation displayed by the Radeon Pro Duo and the pair of Nanos is through the roof.

The line graphs look like something Cthulhu would use to comb his barbels, with spikes over the 50-ms range. Meanwhile, the pair of 980 Tis is wondering what all the fuss is about. PC Perspective's results seem to indicate that CrossFire support could use some work.

Rise of the Tomb Raider is the latest installment in the now decades-long series, and one would have to be blind to say it's not a gorgeous game. At 2560×1440, the Radeon Pro Duo and Nanos put up a decent enough showing, with the usual caveat of less-than-optimal smoothness due to frame time variation. At 4K, however, things take a turn for the worse, and something happens to the Radeon cards.

The initial part of the benchmark is peppered by extreme variation in frame times. We can only guess at what happens, but the cards in the red team behave fine for the rest of the benchmark. The problem is exclusive to CrossFire, however—the single Fury X in PC Perspective's results doesn't exhibit this strange behavior. Neither does the pair of GTX 980 Tis.

Witcher 3 is the most gorgeous PC game out there (there, I said it), and it tends to runs pretty well despite its graphical flourishes. Unless you're using a pair of Radeon GPUs, that is. This is sounding like a skipping CD, but something is up with CrossFire. The Radeon Pro Duo, the set of Nanos, and the Radeon R9 295 X2 all exhibit the same wild variations in the initial stages of benchmarking, akin to what happened with Rise of the Tomb Raider at 4K.

The thing is, it happens with Witcher 3 at both 2560×1400 and 4K resolutions. And yes, we're writing the same thing over and over again: the 980 Ti set takes things in stride even at the highest resolution, with small and steady variation in frame times.

The Radeon Pro Duo isn't being positioned as a solution for gamers, with some good reasons. The card's raw performance is akin to a pair of R9 Nanos. However, a pair of Nanos will currently set you back around $1000, while a Radeon Pro Duo goes for $1500. When comparing power consumption figures, the pair of Nanos and the Pro Duo come up evenly matched, too. However, two R9 Nanos will require two PCIe slots. They're dual-height cards, too, and you'll ideally keep some distance between them, so space is a concern when it comes to setting them up. Having this much horsepower in a single card may be quite appealing to some.

Comments closed
    • DancinJack
    • 3 years ago

    lol wow I come back to these comments and it’s turned into another AMD shillshow and all of the internet is against them. The interwebs really are wondrous.

    • hasseb64
    • 3 years ago

    Next gen GPUs just around the corner, this is just such a waste. Money and heat.

    • Mr Bill
    • 3 years ago

    So, I’d like to hear a pro explain why SLI has such superior latency compared to CF (including CF using a PLX chip).

      • Airmantharp
      • 3 years ago

      Honestly I’d like to see another investigation into this, particularly with DX12 and next-gen cards.

      I won’t claim that SLI is flawless, but it very nearly is; it at least lives up to the promise, which is why I use it. It’d be nice if Crossfire was a credible alternative just for the sake of having that option and for the resulting competition.

    • zzz
    • 3 years ago

    I’ll give AMD a pass on this one, they’re billing this card as a VR dev item, they don’t seem to be pushing it as a gaming card so I won’t give them hate. AMD may suck at marketing but they don’t seem to be pushing this as a gaming card, any benchmark of this card should accordingly include a VR game I guess?

      • derFunkenstein
      • 3 years ago

      That’s the one weird thing about their choices. Having some sort of VR title would have made sense. No Eve Valkyrie, no…whatever other games there are (?). Seems a little weird.

    • odizzido
    • 3 years ago

    What is TRs plan for reviews anyways? I know TR is pretty much a new website as everyone who used to write for it is gone but I was expecting at least something by now.

      • derFunkenstein
      • 3 years ago

      Maybe you missed the whole “AMD didn’t send them one” thing. I know that reading is new for you. 😉

        • odizzido
        • 3 years ago

        I mean in general. I actually never read the duo review linked here as I have no interest in multi-gpu cards.

        Guess I will be waiting till the new GPUs launch for an answer.

    • Computerdr
    • 3 years ago

    I can’t believe nobody has noticed that every title benchmarked here was an Nvidia Gameworks title. Obviously PC perspective is getting compensated to show AMD in a bad light.

    PC Perspective is a major tech website and knew exactly what they were doing when they chose Nvidia friendly titles only. I’m disappointed at Tech Report for making absolutely no mention of this either. Things are starting to smell real stinky around here.

    There are also no DirectX 12 games included where AMD nearly always comes out ahead — AMD has full Tier 3 DX12 support while Nvidia only has T2.

    Perhaps PC Perspective ran out of Gameworks titles to push their agenda and had to cut their “review” short? Who ever reviews a new high-end GPU in only four or five games? And then they have the gall to call the review “THE Radeon Pro Duo review”.

    This is deception at its finest. It’s not a question of if they were paid to publish this, it’s a question of how much. We as gamers who want unbiased, reliable information should be booing these jokers off the damn stage.

      • thereason8286
      • 3 years ago

      I know right. I was just about to post that. All gameworks titles and not a single DX12 game for the sake of just putting it in there cause its clearly going to be the future of most games.

      I have to agree with you completely. Im also quite shocked TR who in my eyes are usually more balanced and fair didn’t notice that. tsk tsk

      Edit: Not only that but i’m surprised that some of regulars here with a lot of knowledge didn’t notice that and instead just accepted it just like that. Heck their comment section is blowing up on how bias they felt the review was..

        • DoomGuy64
        • 3 years ago

        [quote<]Im also quite shocked TR who in my eyes are usually more balanced and fair didn't notice that. tsk tsk[/quote<] Honestly? Have you read their Fury review? That review was so bad that TR's neutral image sank into the ground, and everyone jumped ship all at once. There's barely anyone left of the original crew. It feels better now, but back then TR seemed like they were pulling all their news and ads straight from a nvidia newsfeed, detailing all the gameworks features of various titles, not to mention shield devices. Their Nvidia articles also had 2-3x better writing, so you could tell what the focus was on. You can make statistics support whatever bias you hold, simply by picking games you know perform better on your preferred brand. Concepts like frametimes are abused for a narrative sleight of hand, when you don't pick neutral games to benchmark. The only thing you can really tell from that review is that [i<]those[/i<] games perform worse. It doesn't give you a big picture, and I can't see AMD doing poorly with future titles having the console monopoly, and also the best dx12 support. You'd have a hard time convincing anyone today that the 970 is better than a 390, but that's exactly what happened when those cards were released. IMO that was dishonest reviewers skewing numbers to support their bias. AMD doesn't have marketing money to throw around like Nvidia does, and that causes all sorts of problems in terms of brand perception. Tech sites are quickly becoming as reliable as gaming review sites. If this continues, people will start looking to the community for benchmarks instead of professional reviews. Couple of other points: The April system guide mentions a 960 vs a 380X. The problem with this is that the 380 is the actual competitor to the 960, and TR's 960 pick had 2GB of ram. When you compare a 2/4GB 960 to a 2/4GB 380, the price difference disappears and you are only left with performance. ------ [url<]https://techreport.com/review/29514/amd-sets-a-new-course-for-radeons-with-its-polaris-architecture[/url<] [url<]https://techreport.com/review/30048/exploring-nvidia-pascal-architecture[/url<] Can you spot the difference? The nvidia article has 5 pages, while the polaris article has a short 1 page blurb. This has been par for the course with TR for a long time running. They clearly prefer nvidia, but refuse to admit it outright. Nvidia has also gotten a pass for a lot of things that AMD would have been eaten alive for, and only gets mentioned when those issues hit critical mass. I don't really care to dig through every single example. My only recommendation is to take every review with a grain of salt, and read multiple sites. That's the only sure way to know the truth.

          • ImSpartacus
          • 3 years ago

          Yeah, that Fury review, wowzers. That one is going to be a thorn in tr’s side for a while. I honestly believe that review was the result of time limitations and/or laziness rather than conscious malice/bias.

          But even then, no matter what the reasoning is, you’re still left with a very mediocre & misleading review. Very disappointing.

      • derFunkenstein
      • 3 years ago

      Bias is an incredibly hard sell considering PCP is the only mainstream site with one of these cards, and considering they have in recent months had AMD reps in-studio for live streaming video events.

      DX 12 titles still have full-screen window issues and frame pacing issues with all multi-GPU configs. Avoiding frame time graphs on those titles actually works in AMD’s favor.

        • thereason8286
        • 3 years ago

        I get you but what do you think about the games of choice? I mean all gameworks titles…. Not to mention some very old.. Just seems silly to me. Maybe it’s just me iono.

          • derFunkenstein
          • 3 years ago

          The games are popular and have been in PCPer’s benchmarks for a long time. Some of them are a little older, I guess. I’m not looking for conspiracies at every corner, either, though.

          The big thing is, as zzz said, no VR games. That’s weird.

            • ImSpartacus
            • 3 years ago

            Then could it be said that their “default” choice of benchmark games is a poor selection in general (not just in the context of this particular review)?

            And on top of that, not even trying to do a VR-minded benchmark on this obviously VR-minded card is very suspicious, as mentioned.

            I mean, if AMD wanted a halo gaming card, they could’ve gone the 295X2 route and build a 500W CF-Fury-X-on-a-card. They still wouldn’t’ve “won”, but it would’ve shown an intent to get the most gaming performance possible. Instead, they ensured that the card met the PCIe spec and they consistently tied it to VR.

            There’s no avoiding this card’s failures as a general gaming card, but it just feels weird to do a review without diving into VR. A “review” for the Radeon Pro Duo feels like a fantastic opportunity for a deep dive into the state of VR with the Radeon Pro Duo benchmarks actually taking backstage to some compelling VR commentary.

            • derFunkenstein
            • 3 years ago

            Takes too much effort to find a conspiracy or malice to make their choices suspicious. Much more likely the choices were just kinda dumb.

            • ImSpartacus
            • 3 years ago

            I agree that it’s more likely to be ignorance than conscious “bias”, but even then, that’s a big thing to mess up (no matter the reason).

            I mean, when you’re picking benchmark titles, you’re surely aware that they will be ruthlessly scrutinized. I feel like it’s a decision to not take lightly.

            Even if there’s no bias present, there’s still a massive credibility hit.

            • derFunkenstein
            • 3 years ago

            Definitely, which means that at least part of this falls on AMD. Shouldn’t they insist that their artificially-segmented product at least see some tests in that segment? I mean, surely they gave some sort of guidance. Every company does that in the form of reviewer guides. I imagine the conversation must have gone something like this:

            AMD: “Hey we will give you an exclusive look at a VR-focused product.”
            PCPer: “Cool.”
            AMD: “Could you do at least a few VR tests with it?”
            PCPer: “…”
            AMD: “…”
            PCPer: “It’s a gaming card, right?”
            AMD: “It’ll [i<]run[/i<] games, I gue—" PCPer: "Great!" edit: If they really wanted to get it some VR coverage, they should've given it to Jeff. He's got a Vive and a Rift in the lab.

            • ImSpartacus
            • 3 years ago

            I think there’s two issues at hand and it’s important not to mix them up or “overlap” them together.

            One of the two issues is PCPer’s poor choices in their standard benchmarking selection. I don’t think AMD (or Nvidia) should be held responsible for that. A reviewer like PCPer needs to carefully make those selections and if the selections still have some quirks (which could be unavoidable if you want to benchmark popular games), then they need to plainly & exhaustively explain the quirks in their benchmark games in every single review. Even if you have a crummy pool of benchmark games (via ignorance, bias or whatever reason), if you properly disclose that then no one can fairly accuse you of bias.

            I think the other issue is the lack of reasonable VR benchmarking, as you mentioned. It gets a little murky to discuss AMD’s private discussions, but I agree that the Radeon Pro Duo’s VR use case needs to be in the reviewer guide (and I hope it was). Also, in these unique circumstances, AMD should’ve made an extra effort to “insist” that a VR benchmark be utilized, as you mentioned. However, AMD (or Nvidia) can’t force a reviewer to do something and I think that’s a good thing (for integrity purposes). So ultimately, I put this issue on PCPer’s lap (though they did an admirable job of explaining that it was a VR-minded card).

            • derFunkenstein
            • 3 years ago

            PCPer was a head scratcher of a choice if, in fact, AMD gave them one of these cards. They have not written word one about hands-on VR that I’ve found so it makes zero sense to make them the only outlet to get to test this card.

            The actual tests are fine. These are popular games. AMD gets no breaks on the selection. Either it can run them well or it can’t. If you’re gonna cry over gameworks, that’s not PCPer’s fault. Popular games use gameworks. Big deal.

            What’s been omitted is what sucks. What’s there is ok, I guess.

      • torquer
      • 3 years ago

      You could copy and paste this post into every comment thread on every story since the dawn of time where frame variance is shown to be a weakness in AMD’s solutions.

      If you love the company (AMD) as much as you purport to, you’d encourage them to address their shortcomings rather than whitewashing them or whining about bias anytime any news story or review site finds it to be less than perfect.

      God knows Nvidia is far from perfect but I’m not moaning and wailing every time someone (legitimately) points it out.

        • NeoForever
        • 3 years ago

        You’re right. We could encourage AMD to address the shortcomings (For argument’s sake, nvm the proprietary lockbox that is GameWorks).

        But can’t we ask AMD to fix issues and criticize a review site for a poor selection of games at the same time?
        And why can’t we objectively call it a bias if a review site only selects certain games that are known to perform better for one company (even if that company happens to be AMD)?
        After all that is what “bias” means. I’m not sure intent is important here. Scientifically, it is a biased selection, and hence, the bias should be justified (sometimes it’s perfectly justifiable).

          • derFunkenstein
          • 3 years ago

          Did you read the comment to which that was a response? That’s not at all what computerdr meant. Not a question of if they were paid but how much? Not to mention AMD hasn’t provided samples, let alone retail boxes. Where do you think that card came from? Nvidia? AMD gave them that card and we have to think AMD gave them that card for a reason.

          If anything is a conspiracy, it’s why there were no VR benchmarks. Why is that, AMD? Maybe that’s a better question.

          • torquer
          • 3 years ago

          Because frame pacing is well known to have been a struggle for AMD for ages. Its a fact. Its a fact as much as Nvidia s**tcanning compute performance or struggling with Async. Can it be fixed? Perhaps? Software or hardware issue? Who knows. You’re commenting on a thread on the very site that identified this issue. Hell, the editor of this site now works for AMD!

          The point here is simple. Its a real problem. AMD has had issues with it. They should address them. Whether the issues are pointed out in a way that isn’t inherently perfect from a testing methodology standpoint is irrelevant because the issue is so well known and has been proven in so many ways for so many years it removes the need for pure testing methods every single time its shown to be the case.

          You don’t fix problems by pretending/wishing/excusing them away. You fix problems by facing them head on. Don’t give companies a pass because you love them. It does them, you, and us all a great disservice.

          If you like, to “prove” my objectivity I’d be happy to point out the list of things Nvidia does that sucks, but I’ll need a much longer post and a lot more time 🙂

      • Neutronbeam
      • 3 years ago

      I bet you also hate all those unmarked black helicopters filled with anonymous troops sent out to take over every country and enforce the New World Order too.

      If you’re trolling,okay you win; hands down. If you’re trying to make a legitimate criticism you need a lot more proof points.

      Given that the OWNER OF THIS SITE WORKS FOR AMD, if anything you should be arguing that TR is an AMD fanboy.

      • Jeff Kampman
      • 3 years ago

      Honestly, I have never understood this logic. Whatever you may think of GameWorks, the fact of the matter is that the games PC Perspective tested are wildly popular AAA titles that many people—AMD and Nvidia graphics card owners alike—are going to own and care about as a reference for performance. It’s PCPer’s job—and ours—to use the methods we’ve developed to show how these games perform in the real world, full stop.

      No PC gamer is going to put off playing [i<]The Witcher 3[/i<] or whatever major release on their Radeon just because it performs worse than the GeForce competition in benchmarks on a major tech site. Maybe they already own the graphics card, and they just want to know what to expect. Maybe they're shopping for a new graphics card, and they want to see how their potential choices perform in some wildly popular title. Either way, the general public doesn't care about "GameWorks this" or "Gaming Evolved" that—the performance when the bits hit the silicon in a popular game is all that matters. The furor surrounding our Fury review and [i<]Project Cars[/i<] seemed to spring from a similar vein. Again, we have a popular game that many people like to play. It's entirely plausible that a Radeon owner or shopper would want to know how this popular game would perform on a $550 graphics card, and so we tested it to answer that question. If it didn't perform well on a Radeon, well, that's good for the public to know. I don't think it's an attack on AMD to publish those results and talk about them; the goal is to help the consumer make an informed decision. Hiding data because they make one company or another "look bad" or "lose" is a cop-out, not "fairness." The same is true here. If the Pro Duo—a $1500 graphics card—doesn't perform as well as one might expect in today's most popular games, that's data worth knowing. Whether they're GameWorks titles or whatever is beside the point. Those games are what's available to us in the world we live in today, and it's the context in which people are going to make purchasing decisions about the graphics card. Leaving popular GameWorks titles out of a major graphics card review is like saying "buy a Pro Duo, but don't play GTA V/The Witcher 3/whatever on it." That's so absurd that I won't even go further into it. Leaving titles out of a test suite or including others because of some preconceived notion about the way they might make a card look is the sort of unscientific, match-fixing behavior that you yourself are railing against. Admittedly, PCPer could have tested more games, a broader variety of APIs, whatever. That's always true of any review, but limited time and resources are something we all face in this business, and I could understand why they might have gone for the "heaviest hitters" in the sense of Steam popularity or whatever when they put together their test suite. I'm not sure what the point of this ramble is other than to say that barring some kind of ideological investment in the outcome of a graphics card review, the data PCPer collected looks fine to me. It reinforces what we've known for a long time about CrossFire and the pitfalls that frame-time-based metrics reveal for that multi-GPU setup. Hopefully AMD sees this data and continues to refine their drivers and software to improve the experience for the end-user.

        • f0d
        • 3 years ago

        i would like to second this
        i own a R9-290 and i WANT to know how my/amd cards perform in games like project cars GTA V and the witcher, i want to see how the games that i play perform

        almost all games now are either amd or nvidia sponsored and the only games that are not involved with either company in any way are games that barely anyone plays

        in a perfect world i would say benchmark the 10 most played games that were released in the past year but finding out player numbers for games is impossible so thats a nogo

          • DoomGuy64
          • 3 years ago

          Problem with benchmarking gameworks titles is that it takes AMD longer to optimize for, and when they do that invalidates the old data, while the review in question never gets updated.

          It makes no sense to benchmark those games, especially when PC gamers are more skeptical than console gamers, and do not always buy their games day 1. I certainly don’t, as it’s too expensive, and a total waste to pre-order titles that could potentially be garbage. We don’t know if games are worth buying until after they come out, because game review sites often do not tell the truth about the quality of those games. Any honest hardware site should steer clear of titles like Batman:AK, and minor issue games like PCars need to include a disclaimer. The Fury review was especially problematic because too many of those titles were used, and that gave the appearance of bias.

          We now know the 390 is better than the 970, especially going forward. That wasn’t obvious day one, because games that could measure their actual capabilities weren’t chosen. What got chosen were titles that performed better based on driver optimizations at the time, and not titles that could accurately measure the hardware. Drivers change, hardware does not. I want to know what the hardware is capable of, not which driver is better, although both are important. That’s why I read multiple sites, because you can’t get an accurate representation from reading a single review. Companies can always release a driver that fixes a problematic title, but they can’t change the hardware specs after the fact. Basing a review off titles that are known to have issues is a sure fire way to get accused of bias. Don’t do it. It’s poor journalism, and not representative of a card’s actual ability.

            • Airmantharp
            • 3 years ago

            I still can’t see ignoring popular titles at launch.

            Does this suck for AMD who is usually slower with drivers? Sure, they’ve taken heat for it, and they’ve worked to improve the situation.

            But is it unfair? Hardly. People who buy cards at launch are going to play games with launch drivers, and while improvement over time with driver optimizations is certainly more typical for AMD than it is for Nvidia, it is never a given, nor can it be properly forecast in any meaningful way. All that can really be said at launch is whether the reviewer, subjectively based on their own experience, believes that the hardware is being held back by launch drivers.

            Your issue is this: Nvidia is the bigger dog in the GPU fight, and they’ve innovated across the board to maintain that lead. Some of these innovations, such as helping developers optimize titles, do appear underhanded, but they work- and if AMD wants to compete equally, they’ll have to beat Nvidia at their own game.

            • DoomGuy64
            • 3 years ago

            [quote<]they'll have to beat Nvidia at their own game.[/quote<] Which is exactly what AMD has done with dx12, and consoles having AMD hardware means future games will naturally run better on AMD hardware. Regardless, you're missing the forest for the trees, and advocating an eye for an eye. This isn't a situation where AMD should copy Nvidia, because it would further segment the market. The problem is US, the public turning a blind eye to developers treating their customers poorly. Developers should not be playing favorites, and we should not permit them to do it. Instead of placing the blame solely on AMD or NV for not being able to handle sponsored titles early on, the blame should be on developers who don't optimize equally. They're the real problem in this situation, and we shouldn't give them a free pass by ignoring their part in the debacle. Developers who play this game should be called out on their participation, and not given credibility by including their games in benchmarks. I don't see AMD having a hard time handling dx12, or competing with future games. I fear having a market that fragments to an extreme level where games perform wildly different depending on who sponsored them, and that isn't right. We can't afford to ignore this issue if we want everyone to enjoy their games equally. The answer isn't more unfair tactics, it's equal optimization. The cheating has to stop, so we must call it out before it gets out of control.

            • torquer
            • 3 years ago

            Call it out as much as you like but the market as a whole just doesn’t care. Sorry to break it to you, but if the market cared about all the things we think they should care about we probably wouldn’t have any games at all. There’d be no Ubisoft, there’d be no EA. There’d be no Valve. There’d be no Gameworks, there’d be no DLC, there’d be no proprietary anything, no closed source software. No first person shooters, no games that objectify women or aren’t politically correct. No new games on old engines, no first day patches. You’d get rid of a lot of stuff you hate, but you’d also lose a lot of stuff you love.

            Its great to be passionate about this stuff, but its like with all my highly politicized friends who refuse to do business with any entity that doesn’t conform to their sociopolitical beliefs – in the end it is self defeating because you run out of people to spend your money with. You’re certainly free to do so, but you might as well argue with a street sign for all its worth.

            Vote with your wallet, just don’t be surprised when you find yourself in the minority.

            • Airmantharp
            • 3 years ago

            I don’t think AMD has done anything with DX12- it’s open, after all, and Nvidia’s efficient driver program will no doubt hit DX12 in stride.

            As for the developers caught in between- I don’t blame them. Optimizing their work, often needing that optimization as they seek to differentiate themselves by doing unique things, is absolutely necessary, and it would be fiscally irresponsible for them to turn down help from a graphics vendor.

            This is why AMD has to step up- they have to provide that support to companies so that at the very least developers aren’t haphazardly taking routes that work against the particulars of AMD’s graphics architectures, and so that they have the insight they need to optimize their drivers for launch day. Be that for the launch of new AMD cards or the games themselves.

            It’s up to AMD to take those steps and to prevent the market fracturing that you profess to fear, and it’s in AMD’s best interest to do so just as it’s in developers best interest to accept help where they can get it.

            • beck2448
            • 3 years ago

            Nvidia has a history of waiting until the mainstream titles embrace a new standard before putting all their resources into it. I have no doubt when DX12 is actually retail their performance will be great. Right now DX12 is barely on the sales map.

            • DoomGuy64
            • 3 years ago

            I think we already have an example of how Nvidia is going to handle dx12, and that’s Gears of War.

            DX12 isn’t going to improve Nvidia’s image. Voxel shading drops your framerate to the 30’s even on TI level hardware, async isn’t supported, and low level gameworks isn’t going to work right on AMD hardware. The only logical outcome of nvidia’s dx12 support is a total disaster.

            • beck2448
            • 3 years ago

            The most popular titles get reviewed. Not obscure beta testing titles which are are not retail ready.
            Plus you need Windows 10 that comprises what ,20 %? The whinging about this smacks of desperation.

            • DoomGuy64
            • 3 years ago

            [quote<]Not obscure beta testing titles which are are not retail ready. [/quote<] *Cough* Project Cars *Cough*

            • f0d
            • 3 years ago

            project cars was out and running just fine when it was in reviews as far as i know

            edit: as someone who has been playing it since release i havnt had a single issue with pcars

            • DoomGuy64
            • 3 years ago

            Early Access type game?
            Check.
            Built from the ground up explicitly for Nvidia hardware?
            Check.
            Not a popular title?
            Check. Niche market.
            Missing content?
            Check.

            Status:
            Broken Nvidia tech demo with missing content, full of bugs, and made by developers who have no interest in optimizing for AMD.

            Totally a justifiable benchmark. /sarcasm.

            Yeah, no. It’s obviously not a good benchmark, and the only people defending it are nvidia fanboys, which make its use look all the more illegitimate. Doubling down on a known mistake is the best way to broadcast that you’re biased, and can no longer be trusted. So please, keep it up, because it doesn’t help your cause at all.

            Project Cars is the canary in the coal mine. You can’t advocate for it without signaling that something is wrong with your logic.

            • f0d
            • 3 years ago

            [quote<]Early Access type game? Check. Built from the ground up explicitly for Nvidia hardware? Check. Not a popular title? Check. Niche market. Missing content? Check.[/quote<] early access yes but was not used in video card reviews until it was released as far as i know it was not built from the ground up explicitly for nvidia hardware, its not even a gameworks title and physx is on the cpu just like in games like borderlands (tested myself - physx on cpu or gpu makes zero difference) ok its not a very popular title but it was a popular racing game and reviews usually try and have games of varying types instead of all fps games - same thing happened when they reviewed the R9-290X and they included grid 2 which isnt a popular game but was a popular racing game missing content? you have lost me there, sure they have added content after the release of the game but who doesnt? what was "missing"? [quote<]the only people defending it are nvidia fanboys, which make its use look all the more illegitimate. Doubling down on a known mistake is the best way to broadcast that you're biased, and can no longer be trusted[/quote<] i have a R9 290 here is my 3dmark score [url<]http://www.3dmark.com/fs/8172923[/url<] a post mentioning i have an R9-290 [url<]https://techreport.com/forums/viewtopic.php?f=3&t=117592[/url<] if i am such an nvidia fanboy what the hell am i doing with an amd card?

            • DoomGuy64
            • 3 years ago

            [quote<]early access yes[/quote<] Yup. Still is in a sense. Lot of people complaining about features promised before release not making it into the "final" product. It's still not finished. Devs plan on releasing DLC to cover for that, afaik. edit: Nope. They're releasing Project Cars 2. Possibly to salvage their reputation, at the expense of existing users. [quote<]it was not built from the ground up explicitly for nvidia hardware[/quote<] Yes it was. It's just not an "official" gameworks title. The devs used nvidia tech like physx, and worked directly with nvidia engineers. The game is optimized for Nvidia, and they aren't willing to optimize for AMD unless given special treatment. Which I doubt they'd do anyways, because they'd have to remove physx and replace it with havok. Not going to happen. [quote<]if i am such an nvidia fanboy[/quote<] I'm generalizing. People generally don't apologize for this game unless they're ignoring the controversy surrounding it. There's no way you can salvage this game's poor reputation, outside of the developers promising AMD optimization, bug fixes, and more content. Then actually delivering on those promises. Until then, this game is dead in the water, and there is no saving it. No amount of apologizing can fix the mess that is Project Cars, only results will. There's all kinds of other issues that aren't related to AMD, and without addressing them the existing community will leave for other titles. Apologizing for this game is like giving CPR to a dead horse. It's dead. There is no saving it.

            • ImSpartacus
            • 3 years ago

            Do you have any sources of Hawaii (or other amd chips) improving their relative performance as they age?

            Your post isn’t the first place that I’ve heard this statement and I think it’s plausible, but as you said, I’m a pc gamer and we’re skeptical.

        • ImSpartacus
        • 3 years ago

        I know that the “Project Cars” issue was the most popular issue about the infamous Fury review, but it wasn’t the only issue.

        In my eyes, the biggest issue was that a stock Fury was being compared to a factory overclocked 970 & 980 without that fact being properly communicated to readers. Some context:
        [list<] [*<]The article does not state that the 970 or 980 are overclocked. [/*<][*<]All graphs use the plain term of "GeForce GTX 970" or "GeForce GTX 980", so a layman (such as myself) would probably assume that they are stock. [/*<][*<]The specific model of 970/980 was mentioned only once in the entire article, [/*<][*<]And that one time was at the very bottom of an obscure methodology page. [/*<][*<]The factory-overclocked 970 & 980 clocks are stated at the bottom of the previously mentioned methodology page, but there are no stock clocks for the 970 or 980, so [u<][b<]even a particularly thorough reader that read every single word in the entire review wouldn't be able to figure out that these cards are overclocked unless they had memorized the stock clocks[/b<][/u<]. [/*<] [/list<] The exact discussion where I pointed this issue out is [url=https://techreport.com/discussion/28612/asus-strix-radeon-r9-fury-graphics-card-reviewed?post=921378<]here[/url<], but you can get the gist of it from this comment. Mr. Wasson's very first response was to remind me that the Hawaii cards were also factory-overclocked. I admit that I'm still a little upset that a respected tech blogger's first reaction to learning about a critical systemic documentation error in two popular graphics cards is to boldly announce that it happened in [u<][i<]another[/i<][/u<] two popular graphics cards, as if another two "wrongs" even out the first two "wrongs" just because they occur in competing firms. While I appreciate him sharing that extra info, the irony of the whole situation wasn't lost on me: Even after painstakingly looking at the very methodology table that contained the Hawaii clocks in order to write my comment, I honestly had no idea that they were overclocked until Mr. Wasson mentioned it because [u<][b<]I hadn't memorized Hawaii's clocks[/b<][/u<] and I never thought to look them up (I didn't check anything other than 970/980). Honestly, [i<]that[/i<] moment really drove the issue home for me. In new graphics card reviews after that, the graphs now show the exact make & model of each graphics card, which is good. However: [list<] [*<]TR graphics card reviews still don't display the stock clocks in their methodology page, and [/*<][*<]They don't always mention that some of the graphics cards are overclocked out of the factory. [/*<] [/list<] So, again, unless you memorized the stock clocks of graphics cards or you can recognize model names as being overclocked, then this issue can still occur. It's kinda scary.

          • torquer
          • 3 years ago

          Unless you pay attention to what you’re buying. If you buy a graphics card in a vacuum based SOLELY on a single review from a single site and then regret your purchase because that wasn’t enough info to make you feel warm and fuzzy about said purchase… well you probably deserve it.

          Would you also buy your laundry detergent after watching a single commercial for it? At what point do you all start holding consumers responsible for their own research and buying decisions?

          Or, is it more accurate to say that your noble defense of the hapless consumer is really just window dressing for your butthurt when your chosen brand isn’t presented in the absolute best possible light?

          • DancinJack
          • 3 years ago

          Yeah, you seem like a layman…

            • derFunkenstein
            • 3 years ago

            Yeah I kinda quit reading his comment when I saw “970 and 980 overclock hurr durr” because that’s an old argument. Those cards are retail examples not specialty unobtanium.

            • ImSpartacus
            • 3 years ago

            You’re right. They are common. In fact, the prevalence of overclocked Maxwell cards is one of the reasons why it’s so critical to properly communicate exactly what the overclock is (whether it be high, low or none).

            It’s no secret that Maxwell is a particularly impressive architecture that has a ton of clockspeed headroom. So in [url=http://pcpartpicker.com/parts/video-card/#c=185&sort=a8&page=1<]the 980 realm[/url<], you've got stock reference boards at the bottom for only [url=http://pcpartpicker.com/part/pny-video-card-vcggtx9804xpbcg<]$441[/url<] and for less than thirty bucks more, you can get a [url=http://pcpartpicker.com/part/evga-video-card-04gp43988kr<]fancy EVGA 980[/url<] with a core (i.e. "base") clock that's 15% higher than stock. That's a lot for thirty bucks. With that kind of performance available, I think it's effectively required that a reviewer look at an overclocked 980. However, on the flip side, that tremendous amount of headroom causes a large variance between what a "980" can actually perform like. Therefore, it's critical to clearly state if the card is overclocked and what the overclock is. Honestly, this policy should apply to all cards, but it's especially important for Maxwell and even more important for GM204.

            • ImSpartacus
            • 3 years ago

            I’m flattered, but I have no professional training in this industry and I humbly come to sites like TR or Anandtech to learn.

      • Kretschmer
      • 3 years ago

      If a GPU only performs adequately in certain games, that should be front and center in any review.

        • torquer
        • 3 years ago

        This is a great point. Whether Gameworks is inherently good or bad for the industry, it exists. Removing gameworks titles from benchmarking would leave as unrealistic and irrelevant a sample as exclusively using them.

        The fact of the matter is the broader market probably doesn’t give a flying Chuckula whether a game is Gameworks or not. They want to know how a reasonably broad selection of popular games runs on hardware. Some games will work better with Nvidia, some will work with AMD. It is up to the responsible consumer to spend the time necessary to ensure whatever the hell they buy (which incidentally you should not give a crap about) runs the games they care about well.

        And in the end, that is the whole point of everything. The rest is just a lot of pissing and moaning by people who have way too much interest in the buying practices of other people.

          • BurntMyBacon
          • 3 years ago

          [quote<]Removing gameworks titles from benchmarking would leave as unrealistic and irrelevant a sample as exclusively using them.[/quote<] Agree 100% [quote<]It is up to the responsible consumer to spend the time necessary to ensure whatever the hell they buy (which incidentally you should not give a crap about) runs the games they care about well.[/quote<] This is also true. However, I don't see anything wrong with pointing out issues with reviews in the hopes that you can get better data in the future. That said, I don't think pointing out PCP issues at TR will get any more done to fix the problem than pointing out Tom's Hardware issues at Anandtech. Furthermore, even if the review is biased, you can still get good information from it. For instance, if you assume that "Uber-BIASED" PCP picked out every title in their stack that AMD's new card had an issue with and nothing else, then that's great. Now you have a comprehensive list of things to watch out for and AMD has a comprehensive list of issues that can get fixed that much sooner. There is just one thing you said that bugs me: [quote<]The fact of the matter is the broader market probably doesn't give a flying Chuckula whether a game is Gameworks or not.[/quote<] HOLY CRAP CHUCKULA COMES IN A FLYING VERSION AND I DIDN'T KNOW ABOUT IT!!!

            • torquer
            • 3 years ago

            Agreed all around.

      • travbrad
      • 3 years ago

      There are probably 100x as many people playing DX9 games compared to DX12 games. Why no complaints about the lack of DX9 testing?

    • Mr Bill
    • 3 years ago

    The 3dsMAX FireRender Ray Tracing conclusion looks wrong to me. Author says [quote<]The Radeon Pro Duo shows significant scaling, going from a single GPU to two, improving the render time by almost 60% and, considering that is done with one dual-slot graphics card, that's pretty impressive. [/quote<] Going from 30sec for 1GPU to 19sec for 2GPU is not a 60% improvement in render time. [EDIT]OK I stand corrected.... I see the math now. 100*[(30/19)-1] =100*(1.578947-1) = 58% improvement.[/EDIT] Overall having read the review, I think there is something seriously wrong with CF versus SLI and it seems to me that CF smooths out slightly at higher resolutions. I speculate that higher rendering load at high resolution might require less cross talk between the GPUs. So that might mean that both the PLX switching and the CF have some sort of latency issue that is not present in SLI.

      • protomech
      • 3 years ago

      > Going from 30sec for 1GPU to 19sec for 2GPU is not a 60% improvement in render time.

      Presumably they mean an increase in render rate, not a decrease in render time.

      30 / 19 = 157% of baseline, ie a 57% improvement in render performance.

      Perfectly linear scaling (100% improvement) would be 15 seconds, ie 30 / 15 = 200% of baseline

        • Mr Bill
        • 3 years ago

        OK I stand corrected…. I see the math now. 100*[(30/19)-1] =100*(1.578947-1) = 58% improvement.

    • --k
    • 3 years ago

    AMD in its current state is like the 2nd tier sports team in a major city. You want to support them in spirit, but you wouldn’t want to be caught dead wearing their jersey.

      • Kaleid
      • 3 years ago

      I for one can’t wait for Polaris.

      • Neutronbeam
      • 3 years ago

      THAT was PERFECT.

      • Kretschmer
      • 3 years ago

      I don’t “support” many tech companies, but did switch to the “Red Camp” when Freesync Card + IPS was ~$350 cheaper than GSync Card + IPS. There are more “glitches” or occasional graphical corruption, but it’s infrequent enough that I shrug and soldier on. Hopefully Nvidia will support Freesync by my next GPU upgrade.

      I’d use the analogy of AMD as that neighborhood hole-in-the-wall restaurant with amazing food and cheap prices, but you will find a hair in your dish every once in a while.

    • Airmantharp
    • 3 years ago

    “Something is up with Crossfire”

    I left AMD after trying to live with HD6950 Crossfire- a single GTX670, while objectively slower, played smoother, and SLI has proven to be perceptibly as smooth as a single GPU in my experience with SLI GTX670’s and now SLI GTX970’s. I’m not even entirely convinced that a GTX980 Ti would provide a better experience.

    But to the point, I’d thought AMD had gotten the Crossfire frame inconsistency bugs ironed out recently. Is the evidence in the linked article proof that previous efforts have somehow been undone?

      • EndlessWaves
      • 3 years ago

      The evidence in the linked article shows the 295X2 doing much better, so it may be a quirk of Fiji’s design or HBM causing the issues.

        • DPete27
        • 3 years ago

        Or that there was some serious driver tweaks required to get the 295×2 running smoothly. An effort that was clearly not extrapolated to the Fiji Pro Duo. I guess they’ll be starting back at square 1.

          • Airmantharp
          • 3 years ago

          I guess the question is this: Would these issues apply to say 290X Crossfire?

          Freesync being cheaper and AMD cards being cheaper puts AMD in a good position going into the next generation- but I’m not interested if they bungled Crossfire *again*.

            • RAGEPRO
            • 3 years ago

            In theory, since the 295X2 is simply a pair of 290Xes on a stick, 290X Crossfire performance should be much closer to the 295X2’s reasonably decent showing versus the Pro Duo’s awful one.

        • Mr Bill
        • 3 years ago

        Both use a plx chip to link the two GPU modules. I thought the plx chip was what gave the 295X2 superior performance over a couple of 290’s in crossfire. Really have to wonder what is wrong then in the dual Fiji design that negates that advantage.

      • Redocbew
      • 3 years ago

      If they’re aiming this at developers instead of gamers, then it makes sense the driver would get different treatment also. Many of the tricks and tweaks required in the past to get decent crossfire performance might no longer be applicable. That’s my guess anyway.

      AMD seems to be aware that their drivers need some attention, and I doubt they’d go so far as to make a card specifically for developers if fixing the current drivers was the only thing they had in mind.

        • Airmantharp
        • 3 years ago

        I don’t buy the driver argument here, though, as you can run Radeon drivers. There’s really no explanation for the regression in Crossfire performance and frame delivery consistency here.

    • adampk17
    • 3 years ago

    Is this a review of a review?

      • rxc6
      • 3 years ago

      Inception!!!?

        • chuckula
        • 3 years ago

        It’s like a taco within a taco inside a Taco Bell within a mall that’s inside your review!

          • BurntMyBacon
          • 3 years ago

          What flavor is your taco?
          “TACO FLAVOR!!!”

      • Jeff Kampman
      • 3 years ago

      Something newsworthy happened and we reported on it, just like every other day around here 😉

        • sweatshopking
        • 3 years ago

        YOU DIDN’T MAKE A NEWS STORY THAT I WAS BACK IN CANADA, AND THAT’S FAR MORE INTERESTING (i did appreciate this article) THAN LIKE 100% OF THE STUFF SINCE I LEFT.

          • HERETIC
          • 3 years ago

          WELCOME BACK……………………….

          • derFunkenstein
          • 3 years ago

          who are you, again?

            • sweatshopking
            • 3 years ago

            [url<]https://www.youtube.com/watch?v=U9G18qHPhcM[/url<]

      • ImSpartacus
      • 3 years ago

      Yes and I’m ok with that.

      I wish TR would do more commentary like this. I come here to learn, not to read a glorified press release.

      • Neutronbeam
      • 3 years ago

      Raises the ancient Roman philosophical question on gaming card reviews “Review Libri Qui est ?” — “Who will review the reviewers?”

      • adampk17
      • 3 years ago

      Just to clarify, my comment wasn’t a complaint or meant to be snarky. I’m a daily reader of TR and I just don’t recall seeing many, or possibly any, posts quite like this in the past.

      • just brew it!
      • 3 years ago

      Is this a review of a review of a review?

        • cphite
        • 3 years ago

        OPTION (MAXRECURSION 200)

    • SoundChaos
    • 3 years ago

    I honestly cant make anything out of those FPS graphs, not colorblind friendly at all 🙁

      • Voldenuit
      • 3 years ago

      I’m not colorblind and it weren’t easy, either.

      • just brew it!
      • 3 years ago

      Given how common color blindness is, it infuriates me how much red/green is used to indicate completely opposite conditions. At least they’ve figured it out for traffic signals — the red contains enough orange, and the green contains enough blue, that most people with red/green color blindness (the most common type) can tell them apart.

      Pure red/green on computer displays and individual red/green LED indicators are another story. I have a very hard time with those.

        • tsk
        • 3 years ago

        I have perfect vision, yet I find those graphs horrible as well.

        • DPete27
        • 3 years ago

        Well, when red and green resemble each company’s main color….it’s only logical to choose those colors in your graphs.

      • NeoForever
      • 3 years ago

      Their graph’s have always been crap. Even after so much feedback.
      Idk why it is so hard to post crisp pictures on a website. It’s only 2016.

    • chuckula
    • 3 years ago

    Given the results I’d be more inclined to get the GTX-980Ti in SLI and save $300.

      • derFunkenstein
      • 3 years ago

      Considering that the Nano was basically a GeForce GTX 980, a Pro Duo was never gonna really take the performance crown when dual cards is possible.

        • ImSpartacus
        • 3 years ago

        Yeah, the moment we knew the TDP, we all knew that it wasn’t going to be a proper “Fury X2” like the 295X2 was for the 290X.

      • SomeOtherGeek
      • 3 years ago

      That is exactly my thought. AND it would have worked better too. WTF?!?

    • DancinJack
    • 3 years ago

    So, can someone explain to me the purpose of this card? What advantage does it get you?

      • sweatshopking
      • 3 years ago

      dual fiji? it gets you dual fiji. why you’d want that is not clear at all.

        • SomeOtherGeek
        • 3 years ago

        Aaaaaaaaaagh! CAPS filter applied again.

        • Ninjitsu
        • 3 years ago

        O/

        (that’s supposed to be a hand-wave lol)

      • Ninjitsu
      • 3 years ago

      A lighter wallet for more agility

      • Paine
      • 3 years ago

      Duo. Twice the power allows you to watch your movies twice as fast!

      • just brew it!
      • 3 years ago

      The conclusion of the linked article sums it up quite nicely:

      [quote<]AMD started out our most recent conversations about the Radeon Pro Duo by telling us that it wasn’t aimed at the PC gaming market. And I agree – it would be very hard to recommend this card to any PC gamer, including those with a limitless budget. ... For professionals looking to get an AMD multi-GPU configuration in their machine for development and offline rendering, the Radeon Pro Duo is an ideal solution.[/quote<] It is aimed at a very narrow niche market. And that market is not gaming.

        • MathMan
        • 3 years ago

        The Pro Duo is an ideal solution for a market that doesn’t really exist. One wonders if the publicity and buzz that surrounds a card like this is worth the engineering expense…

          • just brew it!
          • 3 years ago

          I did say it was a [i<]very narrow[/i<] niche... 😉

            • MathMan
            • 3 years ago

            There are already reports on some forums from people who bought one, so there is an existence proof!

        • Krogoth
        • 3 years ago

        It is geared towards the same demographic that opts for Titan X (marginally faster than 980Ti). Gamers who want the absolute best no matter the cost.

        Professionals have no reason to get Radeon Pro since it lacks several features that FirePro counterpart would have (10-bit, “certified” drivers full DP, ECC etc) just like the aforementioned Titan X.

      • Krogoth
      • 3 years ago

      Having the fastest single-card GPU gaming solution on the market. That’s about it.

      It the same tier as Titan X and other crazy dual-GPU GPU gaming solutions.

Pin It on Pinterest

Share This