History and Specifications
AMD has launched what it calls the “world’s fastest graphics card”. Is it?
The Radeon Pro Duo had an interesting history. Originally shown as an unbranded, dual-GPU PCB during E3 2015, which took place last June, AMD touted it as the ultimate graphics card for both gamers and professionals. At that time, the company thought that an October launch was feasible, but that clearly didn’t work out. When pressed for information in the Oct/Nov timeframe, AMD said that they had delayed the product into Q2 2016 to better correlate with the launch of the VR systems from Oculus and HTC/Valve.
During a GDC press event in March, AMD finally unveiled the Radeon Pro Duo brand, but they were also walking back on the idea of the dual-Fiji beast being aimed at the gaming crowd, even partially. Instead, the company talked up the benefits for game developers and content creators, such as its 8192 stream processors for offline rendering, or even to aid game devs in the implementation and improvement of multi-GPU for upcoming games.
Anyone that pays attention to the graphics card market can see why AMD would make the positional shift with the Radeon Pro Duo. The Fiji architecture is on the way out, with Polaris due out in June by AMD’s own proclamation. At $1500, the Radeon Pro Duo will be a stark contrast to the prices of the Polaris GPUs this summer, and it is well above any NVIDIA-priced part in the GeForce line. And, though CrossFire has made drastic improvements over the last several years thanks to new testing techniques, the ecosystem for multi-GPU is going through a major shift with both DX12 and VR bearing down on it.
So yes, the Radeon Pro Duo has both RADEON and PRO right there in the name. What’s a respectable PC Perspective graphics reviewer supposed to do with a card like that if it finds its way into your office? Test it of course! I’ll take a look at a handful of recent games as well as a new feature that AMD has integrated with 3DS Max called FireRender to showcase some of the professional chops of the new card.
Radeon Pro Duo Details
The information provided here is an overview of the specifications and design of the card itself. If you read over our preview story already, there isn’t much new here other than a couple of photos we took in-house. If you are ready to jump right to the test results, feel free to do so!
The design of the card follows the same industrial design as the reference designs of the Radeon Fury X, and integrates a dual-pump cooler and external fan/radiator to keep both GPUs running cool.
The 8GB of HBM (high bandwidth memory) on the card is split between the two Fiji XT GPUs on the card, just like other multi-GPU options on the market. The 350 watts power draw mark is exceptionally high, exceeded only by AMD’s previous dual-GPU beast, the Radeon 295X2 that used 500+ watts and the NVIDIA GeForce GTX Titan Z that draws 375 watts!
Here is the specification breakdown of the Radeon Pro Duo. The card has 8192 total stream processors and 128 Compute Units, split evenly between the two GPUs. You are getting two full Fiji XT GPUs in this card, an impressive feat made possible in part by the use of High Bandwidth Memory and its smaller physical footprint.
| Radeon Pro Duo | R9 Nano | R9 Fury | R9 Fury X | GTX 980 Ti | TITAN X | GTX 980 | R9 290X | |
|---|---|---|---|---|---|---|---|---|
| GPU | Fiji XT x 2 | Fiji XT | Fiji Pro | Fiji XT | GM200 | GM200 | GM204 | Hawaii XT |
| GPU Cores | 8192 | 4096 | 3584 | 4096 | 2816 | 3072 | 2048 | 2816 |
| Rated Clock | up to 1000 MHz | up to 1000 MHz | 1000 MHz | 1050 MHz | 1000 MHz | 1000 MHz | 1126 MHz | 1000 MHz |
| Texture Units | 512 | 256 | 224 | 256 | 176 | 192 | 128 | 176 |
| ROP Units | 128 | 64 | 64 | 64 | 96 | 96 | 64 | 64 |
| Memory | 8GB (4GB x 2) | 4GB | 4GB | 4GB | 6GB | 12GB | 4GB | 4GB |
| Memory Clock | 500 MHz | 500 MHz | 500 MHz | 500 MHz | 7000 MHz | 7000 MHz | 7000 MHz | 5000 MHz |
| Memory Interface | 4096-bit (HMB) x 2 | 4096-bit (HBM) | 4096-bit (HBM) | 4096-bit (HBM) | 384-bit | 384-bit | 256-bit | 512-bit |
| Memory Bandwidth | 1024 GB/s | 512 GB/s | 512 GB/s | 512 GB/s | 336 GB/s | 336 GB/s | 224 GB/s | 320 GB/s |
| TDP | 350 watts | 175 watts | 275 watts | 275 watts | 250 watts | 250 watts | 165 watts | 290 watts |
| Peak Compute | 16.38 TFLOPS | 8.19 TFLOPS | 7.20 TFLOPS | 8.60 TFLOPS | 5.63 TFLOPS | 6.14 TFLOPS | 4.61 TFLOPS | 5.63 TFLOPS |
| Transistor Count | 8.9B x 2 | 8.9B | 8.9B | 8.9B | 8.0B | 8.0B | 5.2B | 6.2B |
| Process Tech | 28nm | 28nm | 28nm | 28nm | 28nm | 28nm | 28nm | 28nm |
| MSRP (current) | $1499 | $499 | $549 | $649 | $649 | $999 | $499 | $329 |
The Radeon Pro Duo has a rated clock speed of up to 1000 MHz. That’s the same clock speed as the R9 Fury and the rated “up to” frequency on the R9 Nano. It’s worth noting that we did see a handful of instances where the R9 Nano’s power limiting capability resulted in some extremely variable clock speeds in practice. AMD recently added a feature to its Crimson driver to disable power metering on the Nano, at the expense of more power draw, and I would assume the same option would work for the Pro Duo.
The rest of the specs are self-explanatory – they are double everything of a single Fiji GPU. The card will require three 8-pin power connectors, so you’ll want a beefy PSU to power it. In theory, the card COULD pull as much as 525 watts (150 watts from each 8-pin connector, 75 watts over the PCI Express bus).
AMD is definitely directing the Radeon Pro Duo towards professionals and creators, for several reasons. In terms of raw compute power, there isn’t a GPU on the market that will be able to match what the Pro Duo can do. For developers looking to have access to more GPU horsepower, the price of $1500 will be more than bearable and will give a pathway to really start diving into multi-GPU scaling integration for VR and DX12. AMD even calls out its FireRender technology, meant to help software developers integrate a rendering path for third-party applications.
But calling yourself out as the “world’s fastest graphics card” also means you are putting yourself squarely in the sights of PC gamers. At the Capsaicin event, AMD said the card was built for "creators that game and gamers that create." AMD claims the Radeon Pro Duo is 1.5x the performance of the GeForce GTX Titan X from NVIDIA and 1.3x the performance of its own Radeon R9 295X2.
Obviously the problem with the Radeon Pro Duo for gaming is that it depends on multi-GPU scaling for it reach its potential. The Titan X is a single GPU card and thus NVIDIA has much less trouble getting peak performance. AMD depends on CrossFire scaling to get peak performance (and the rated 16 TFLOPS) for any single game. For both NVIDIA and AMD, that can be a difficult process, and is a headache we have always discussed when looking at multi-GPU setups, whether they be on a single card or multiple.
The build of the Radeon Pro Duo is impressive. Much like the Fury X that was released last year, the RPD design is both sleek and classy, representing the gaming market better than any previous AMD reference products we have tested.
The Radeon Pro Duo is heavy though – so be careful if you are shipping a system with one installed. Mounting the water cooling radiator is a bit easier thanks to the extended tubing compared to the Fury X, which is a nice change. The red Radeon branding along the top of the card remains part of the design as well, and it helps you stand out if you are building the PC inside a windowed case.
Even better – there is NO noticeable pump noise or coil whine from the card! Unlike the Fury X sample we got on day one, where I cringe starting it up each and every time we have to do testing, the Radeon Pro Duo appears to have been fitted with higher quality pumps and electrical components.
Even with async titles this
Even with async titles this card makes little sense because amds own cards can beat it at less money. But now we know bias when its clearly staring us in the face!
pcper doesnt even try to be
pcper doesnt even try to be objective anymore lol
no dx12 testing
only benching on gameworks titles
only benching on titles that is known that amd is behind
dont bench any of the recent games on dx11
good old pcper i dont even wanna think about what they will think to do when polaris will come out
im getting sick of these
im getting sick of these reviews all over the web that are clearly paid for by nvidia
amd has the superior products yet nvidia pays off all the websites to make amd look bad
you couldnt pick a more nvidia biased set of games if you tried – they are all gameworks games and no dx12 to be seen
nobody cares about dx11 anymore
the few unbiased reviews around all show the furyx is a much better card than 980ti when tests are not biased against amd and i bet when the unbiased reviews for radeon pro duo come out all these supposed issues with the radeon pro duo in this review will just disappear
WAKE UP nvidia owners.!
you are being sold a lesser product for a higher price and you are being sucked in to what the paid for websites and the BS marketing are brainwashing you to buy, if you use your brain you will see amd are the best in every situation and every price point
Yes Confirmed
Nvidia paid PC
Yes Confirmed
Nvidia paid PC perspective
i do not even have to tell you as its so damn obvious
I’m laughing pretty hard
I'm laughing pretty hard here. There are some real technical limitations right now to benchmark VR and DX12 that simply cannot be overcome at this time. Remember that magic VR stuff that Oculus does with frame present/shifting to allow a smoother experience for the user? How exactly do you measure frame results when the Oculus software actively changes and adjusts presentation?
Throw in DX12 and how it again has changed how frames are displayed so that FRAPS and the frame rating tech are not accurate. The benchmark for Ashes has also been shown to be inaccurate with what it reports vs. what is displayed and experienced. Would you rather we present misinformation?
Too bad Ryan didn't attempt a title like DiRT Rally though. I do know how much time he spent trying to get this all done before the weekend hit, so can't blame him for not testing everything under the sun. I like AMD stuff, but that $1500 price tag on this card just takes it out of the reach of any sane gamer. Get 2 x Nanos instead.
What about the Pascal and
What about the Pascal and Polaris GPUs any ETA on the benchmarking samples or any non NDA violating information about any engineering/review samples. Hell even some NDA violating leaks from those usual sources.
The DX12/Vulkan and VR benchmarking software stack needs to catch up to the new graphics APIs/driver and VR technology, and that is a story in itself. Looks like the testing tools are going to take time with all the hardware/graphics API changes for GPU’s and VR coming online over the next few months/years!
Haven’t heard nor seen any
Haven't heard nor seen any real leaks about the upcoming products from NV and AMD. They are going to be very interesting from both sides, as they are taking some drastically different philosophies in design and production. You are correct though that the tools in place to accurately measure performance under these new circumstances are just not there. Not only that, but support for Vulkan and DX12 from the GPU guys is also very new and very unoptimized. Things will certainly start to clear up this summer though!
It might be time to clone
It might be time to clone what Digital Foundry does – measure everything externally. Their setup might be insanely expensive but we need to be looking at the video output rather than what software might or might not be telling us.
Ryan had time to test 4
Ryan had time to test 4 games. Choose 4 that were under GameWorks programs especially given how out spoken Nvidia is about helping game developers work since day 1 with the program.
This kind of credibility and excuses of limited timing is laughable to any serious on looker that follows GPU hardware. He could have 2 GameWorks games and 2 AMD games to be balance but that’s still too small of a game pool.
Did Ryan not think or did he just say screw it which seems to be the on going sentiment for the past few years not only here but around the community.
PCPerspective is becoming more of a bulletin board news posting of other sites. More and more of the news here is linked to others doing the work. The time limit seams to be self induced. Delay the article to have more time and include a more balance un-bias set of games. Instead of rushing it out to be first with only GameWorks titles.
Some of us have lives outside
Some of us have lives outside of doing reviews. Ryan has family stuff to do and can't test every game under the sun. As it is the few games tested took two days of straight testing and over 1TB of frame rating output video to process.
He could have just skipped this one / not bothered trying to get a card in to test, which would have meant zero frame pacing results as other sites are either incapable or don't bother. You're welcome.
You could clone 10 Ryans and
You could clone 10 Ryans and that still would no be enough manpower until the Benchmarking software, the graphics API software, and the driver software is there and tweaked with the proper amount of tasting/method calls built into the various software/testing/driver code/API code base. Those benchmarking suits of software require specific API code hooks/method calls in the close to the metal parts of the code base to be able to get the data to properly analyze the frame variance/frame rates/other statistics.
So until the thousands of programmers on all these different parts of the CPU and GPU software/driver/API/Gaming Engine/Gaming engine SDK/OS device driver model/other software stacks get their code bases fully up and running with the proper amount of Testing/Debugging method calls built-in, no amount of Ryans X 10**10 is going to be able to do any testing.
edit: tasting
to testing
damn
edit: tasting
to testing
damn agressive spell checker, and my blindness!
So your response is be
So your response is be grateful and thankful for our bias review.
Stop making sorry excuses. He could have delayed it and picked better games instead of being so bias choosing games under the GameWorks program.
He got the card from other then AMD so he had plenty of time. His restriction and limitations were put on by himself.
Other sites have 3 to 4 times more games they test with just a many more card option as comparison.
Thanks again Allyn for showing the typical response of PCPerspective when the basics are even questioned.
Allyn wants us to think Ryan
Allyn wants us to think Ryan is the only one in the tech review community that does testing and reviews that has a life.
DiRT Rally would have been a
DiRT Rally would have been a much better choice. So would several other more recent DX11 titles. It’s not the lack of DX12 stuff that bothers me so much as the ridiculous decision to pick four of the worst offending Gameworks titles – one, GTA V, being nearly a 3-year-old game at this point (9/2013). The problem is, if you look at the theoretical inverse of this review, it’s like benchmarking a new Nvidia card using only AotS, Hitman, Quantum Break, and the DX12 enabled 3Dmark.
Much like everyone else here, I follow this stuff obsessively. It’s a hobby. If tasked with choosing four titles to guarantee an Nvidia win, at least 3 of the four Ryan picked would have certainly made my top five. That fact alone is rather disconcerting.
Forget this Nvidia paid rag!
Forget this Nvidia paid rag!
Well I will never be using
Well I will never be using this website for reference when it comes to graphics card benchmarks again, for nvidia or amd, as this is such a bias review that these numbers look fake, yet alone are 4 of the heaviest nvidia titles on the market. I understand nvidia pays you, but maybe its time they see they need to actually do something with their products.
This article needs to either
This article needs to either be:
-Completely taken down
OR
-Heavily revised with more titles.
Remember in 2006 when Intel had no way of competing with AMD but to threaten OEMs to stop using their chips? 10 years later, now Nvidia is pulling the same stunts. Or, PC Perspective themselves are biased and should no longer be seen as a credible source.
Ryan Shrout, you know exactly what you were doing when you made this “review” specifically with Nvidia-biased titles. You knew EXACTLY which titles to choose to push your agenda.
And to all my fellow gamers reading this: we deserve better.
All games were nvidia twimtbp
All games were nvidia twimtbp
Guys I’m not defending PcPer
Guys I’m not defending PcPer here but there are other factors than just the Nvidia titles. Yes they were all Nvidia titles and he could thrown in at least one neutral or Amd title but the main problem is the cards themselves. They are two R9 Nanos. Last time I checked Nanos will not beat a 980 ti in anything. Also the card was throttling itself to stay within the power envelope. I think it reached 1ghz once, but most of the time stayed at 950mhz or below. That will cause a lot of frame drops in itself. What Ryan could have done is raised the power up 50 percent so that the cards stayed near 1ghz.
This would have put it on par with Nvidia’s boost technology a little, but as it stands he just did out of the box testing to see what it will do.
All great games that people
All great games that people have played or should play, should they instead focus on games no one cares about just because they are DX12? Both Hitman and Ashes are easy to pass on while the games in this review will go down as some of the best PC gaming has to offer.
Who would buy this? Cross
Who would buy this? Cross fire and wstercooling? Much slower than 2 980tis and stuttering? Is there a single DX 12 game that is retail, not beta? This is not going to be a big seller.
Amd knows this and that is
Amd knows this and that is why they said that they were targeting this product at devz and vr devs, hence the high price tag. But certain people are buying this product and it is selling somewhat. I suspect the price will drop quite a bit in a few months.
They are buying it to get the
They are buying it to get the pro drivers and develop for markets/applications that need to be tested with the pro drivers. So if you were to get a FirePro branded version of this card it would probably cost more than twice what this SKUs costs!
Look at how AMD is marketing this SKU and that’s who will be using this SKU, to any developers that do not have the money for the FirePro versions this is a great deal. And Gamers, wait for other gaming reviews to come out on this SKU that are also using the gaming oriented graphics drivers.
For the developers VR/Others the option to be able to install the professional grapics drivers on this SKUs is a great deal for some cash strapped operations developing applications that are targeting the FirePro branded SKU market that uses on the professional graphics drivers. Pro graphics drivers are designed for accuracy and not Frames Per Second, the gaming drivers use differnet math/other libraries that are quick and dirty and designed for speed not accuracy.
Focus on Polaris/Pascal as that’s where both AMD/Nvidia have shifted their resources, the benchmarks on the Radeon Pro Duo will have to be revisited and go read other reviews on other websites to get more and different benchmarks!
BUT, you know that there are probably testing/review Polaris/Pascal meetings and learning that has to be done by the online press under NDA so that the Polaris/Pascal reviews can be done and revealed at the proper time. The learning curve on the new technology is hard for everyone, go read the AMD and Nvidia white papers on their new GPU technologies along with the white papers on the DX12/Vulkan APIs(both still being Tweaked and fully adopted), it’s not going to be this year before the dust settles and a clear winner can be crowned!
IT takes TIME!
If you decide that this is
If you decide that this is much slower thatn 2 980Tis and with much more stuttering based on two nvidia games, then you deserve them ripping you off selling their shit to you.
The amount of contempt shown
The amount of contempt shown by PCPER save for Walrath, for the overwhelming backlash for this article is disconcerting. Guys you should listen to your readers. Because in this case we’re right. Regardless of whether or not this card is a good buy, your review of it is shit. Listen to your readers and be willing to admit when you’ve done a disservice.
I AM THE WALRATH!
Actually,
I AM THE WALRATH!
Actually, it isn't contempt you are seeing. I think it is exasperation. Everytime we say something critical about one company or another, we get the same response. Shill, paid reviewer, bias… or as we say it, "SO BIOSED!"
Is this the perfect article? No, but it is a good start. I think it exposes some of the weaknesses of the card. I think it also shows exactly why AMD released it when they did and at what price and their marketing push for "developer, content creator, etc." rather than gamers with a top end budget.
Win 10 and DX12 have been a rough patch for everyone involved. I know I was pretty critical of NV in a Sept. through Jan. timeframe for their driver and SLI support. They powered through it and now *most* titles work fine, but they have their corner cases as well where it just falls down. AMD has improved their driver support by leaps and bounds over the past two years, but CrossFire is still a sticking point on more games than what SLI has problems with. Keeping the pressure on these companies, no matter what side you are on, will lead to improvements. Blind support and excuse making will only lead to stagnation because they have no reason to improve the current experience.
Please for the love of all
Please for the love of all that is fair and just! Do NOT leave the Vulkan graphics API out of any discussion going froward! The Graphics APIs(DX12 and Vulkan) are brand spanking New and the testing/benchmarking software stack has not been updated to be able to use these and the other (M$’s UWP nonsense) and other windows 10 DDM/other Nonsense as well as Steam OS based gaming options!
There are gamers and others looking to go Vulkan/other OS in the time up to and after 2020, so do not froget that before and after 2020 there will be gaming on other OSs using the Vulkan graphics API. Even under windows 7, If M$ does not poison pill Vulkan under windows 7, there may be some gamers doing that for their gaming once 2020 arrives(Windows 7’s EOL) and windows 7 has to be safely locked down inside a VM to run some legacy games OpenGL/OpenGL calls wrapped in Vulkan wrapper code/API calls.
Thanks for the cool-headed
Thanks for the cool-headed response, Mr. Walrath. I re-read my post the morning after and winced at calling the review shit. It is not. I take that back, especially after you stated that you all were “keeping the pressure on these companies….” I’ve been considering negative press from only the consumer side, thinking that if we want a competitive marketplace, poor ole struggling AMD can’t afford this kind of lopsided review. I never considered these companies are also very interested in how their products are reviewed. Calling them out on their shortcomings, even if that means a review with games seemingly stacked in Nvidia’s favor, should only solidify their resolve to improve. Having said that, this review was suspectly void of the areas where this card would be superior to its Nvidia counterpart. A few of the things the card is doing right should be mentioned.
The thing is, anyone could
The thing is, anyone could have anticipated the results we see here. This card is two Nanos in CrossFire, period. We know the warts with CF and shouldn’t be surprised at the benchmarks. It might have been better to just concentrate on pro software results only at this stage.
You bunch on whinging
You bunch on whinging wankers.
Look at all these butthurt
Look at all these butthurt (paid) AMD fanboys.
HAHA!
Yay, another nvidia
Yay, another nvidia infomercial.
And this time the “tester”, who admitted recently that is is totally biased toward nVidia in his infomercial about drivers, even comes here to *sigh* about his audience.
Pcper, best “tech” site ever.
Ryan is usually better than
Ryan is usually better than this. But good people can also make mistakes. Lets move on.
PS: Why not add a few more games, so everybody will be happy and it will portray a better picture of the product overall.
No way to make everyone happy
No way to make everyone happy unfortunately. We are investigating new games to benchmark but it isn't as easy as just benchmarking commentors favourite games. There are qualifications which need to be asssessed such as longevity and the ability to framerate it. There are also those like Far Cry Primal which are not different enough from their predecessors to provive worthwhile data.
Like just about any other job, it is a hell of a lot harder than it looks to an observer.
Thanks for the rational comment! 😉
Wow! a $1500 turd. So, buy
Wow! a $1500 turd. So, buy Nvidia if you want the best experience or buy AMD if you like shitting your pants over arbitrary nonsense written on a sheet of paper.
First setup with 2 GPU
First setup with 2 GPU working with HMD and zero VR test?
I think this article is a
I think this article is a fair evaluation of the card.
I would argue that there is
I would argue that there is still a case to buy this card if you’re a gamer, but only in a very specific scenario. If you’re building an ITX gaming system and are looking for the absolute most performance you can wring out of the form factor, this is the card to do it. Having only one expansion slot means that the repeated suggestion of getting two single GPU cards just isn’t an option. Considering how popular ITX is recently, I’m surprised this point wasn’t given more attention.
much waaawwww such typing
much waaawwww such typing