GPUBoss Review Our evaluation of Pro Duo vs 980 Ti among Desktop GPUs

Gaming

Battlefield 3, Battlefield 4, Bioshock Infinite and 21 more

Graphics

T-Rex, Manhattan, Cloud Gate Factor, Sky Diver Factor and 1 more

Computing

Face Detection, Ocean Surface Simulation and 3 more

Performance per Watt

Battlefield 3, Battlefield 4, Bioshock Infinite and 32 more

Value

Battlefield 3, Battlefield 4, Bioshock Infinite and 32 more

Noise and Power

TDP, Idle Power Consumption, Load Power Consumption and 2 more

8.1

Overall Score

Winner
Nvidia GeForce GTX 980 Ti 

GPUBoss recommends the Nvidia GeForce GTX 980 Ti  based on its benchmarks, compute performance and noise and power.

See full details

Cast your vote Do you agree or disagree with GPUBoss?

Thanks for adding your opinion. Follow us on Facebook to stay up to date with the latest news!
VS

Differences What are the advantages of each

Front view of Radeon Pro Duo

Reasons to consider the
AMD Radeon Pro Duo

Report a correction
Much wider memory bus 8,192 bit vs 384 bit More than 21.2x wider memory bus
Is dual GPU Yes vs No About half of graphics cards are dual GPU
Much better floating-point performance 16,384 GFLOPS vs 5,632 GFLOPS Around 3x better floating-point performance
Much higher texture rate 512 GTexel/s vs 176 GTexel/s Around 3x higher texture rate
Many more shading units 8,192 vs 2,816 5376 more shading units
Many more texture mapping units 512 vs 176 336 more texture mapping units
More memory 8,192 MB vs 6,144 MB Around 35% more memory
Significantly higher pixel rate 128 GPixel/s vs 96 GPixel/s Around 35% higher pixel rate
Significantly more render output processors 128 vs 96 32 more render output processors
Significantly better cloud gate factor score 27.47 vs 23.42 More than 15% better cloud gate factor score
Better ocean surface simulation score 1,917.78 frames/s vs 1,735.24 frames/s More than 10% better ocean surface simulation score
Front view of GeForce GTX 980 Ti

Reasons to consider the
Nvidia GeForce GTX 980 Ti

Report a correction
Much higher effective memory clock speed 7,012 MHz vs 1,000 MHz More than 7x higher effective memory clock speed
Better PassMark score 11,318 vs 9,484 Around 20% better PassMark score
Much higher memory clock speed 1,753 MHz vs 500 MHz More than 3.5x higher memory clock speed
Better PassMark direct compute score 6,435 vs 5,267 More than 20% better PassMark direct compute score
Significantly lower TDP 250W vs 350W Around 30% lower TDP

Benchmarks Real world tests of Radeon Pro Duo vs GeForce GTX 980 Ti

Bitcoin mining Data courtesy CompuBench

Radeon Pro Duo
593.55 mHash/s
GeForce GTX 980 Ti
537.83 mHash/s

Face detection Data courtesy CompuBench

Radeon Pro Duo
141.99 mPixels/s
GeForce GTX 980 Ti
145.53 mPixels/s

Ocean surface simulation Data courtesy CompuBench

Radeon Pro Duo
1,917.78 frames/s
GeForce GTX 980 Ti
1,735.24 frames/s

T-Rex (GFXBench 3.0) Data courtesy CompuBench

Manhattan (GFXBench 3.0) Data courtesy CompuBench

Fire Strike Factor Data courtesy FutureMark

Sky Diver Factor Data courtesy FutureMark

Cloud Gate Factor Data courtesy FutureMark

Comments

Showing 24 comments.
Did not expect these many responses...
Hmmm I searched for a comparison of the Pro Duo and the ti and what do I see? the website selling NVidia's by including the 1070. And of course I am willing to bet that the benchmarks have not been updated since the into of the Pro Duo. Not to mention that DX12 performance is not factored in. In other words, gpuboss is now just another 2 bit website pimping for NVidia. Good grief.
Never seen an Nvidia product that has the same amount of VRAM that scales better than an AMD product. Want proof? 8K resolution? Since when Nvidia cards support more than 3 monitors to scale upto 4x 4K monitors (relatively 8K) to test? super-res downscaling? I doubt they support that kind of supper-res. When things drop below playable frame rate, it doesn't matter anymore. If I knew a way to block you, ..... that would be nice. :) Oh I found it. Blocked. :)
LMFAO More NON STATEMENTS!! You are hilarious mate! http://techreport.com/blog/28800/how-much-video-memory-is-enough Funny how when a Fury X, and ALL other GPUs are pushed to the absolute limits at 4k to 8k resolutions, the R9 Fury X scales WORSE than a GTX 980! Again, you come here just to call me "kid" you puerile and infantile imbecile. You did not even READ CORRECTLY! That is not a test of "How much VRAM is in use" it is a test of FUNCTIONAL VRAM BANDWIDTH and how well Nvidia vs AMD scales at higher resolutions - spoiler alert - the 980 does FAR better than the Fury X in all metrics used to test memory subsystems in this article! "For whatever reason, a 4GB memory capacity limit appears to create more problems for the Fury X than it does for the GTX 980. As a result, the GTX 980 matches the performance of the much pricier Fury X at 5760x3240 and outdoes it at 33 megapixels." RIGHT out of the quoted article you illiterate! And those graphs? Yeah ALL IN FAVOR OF NVIDIA! The effective bandwidth of the Fury X in this test was only 387 GB/sec! When it is RATED at 512! Whereas the 980 Ti managed 378n GB/sec despite being rated for just 337! You have got to be the most persistent imbecile I have ever met! Cannot even read benchmarks properly and makes statements that are categorically false! "the only thing I see is that GTX 970/980/Ti use more VRAM than equivalent AMD 290/X/Fury at the same settings" Funny that! Seeing as the Titan X with 12 GB and 980 Ti with 6 got IDENTICAL scores all the way to the insane 7680x4320 resolution! At which point the GTX 980 regular beats ALL AMD CARDS! Then towards the bottom, the Fury X has the WORST frame time of ANY of the tested cards! Literally beaten by a 780 Ti! Again, Good game! Play again, this is getting hilariously fun! Ahh the tech illiterate fan boys trying to argue with someone who actually works with all of these things - please grasp at more straws, call me a kid! Does change your lack of grey matter - and inability to read a tech review properly!
Yeah, "While the Fury X does use some form of memory compression it is not NEARLY as effective as what Maxwell accomplishes" The only thing I see is that GTX 970/980/Ti use more VRAM than equivalent AMD 290/X/Fury at the same settings. What kind of "more effective" memory compression algorithm could use more VRAM instead of less? I care less about fanboy reports. Look at the numbers, kid. Cheers~
That statement is both correct and incorrect. Yes at 512 GB/sec the Fury X with its 500 MHz (not 1 GHz) HBM-1 does indeed outpace a stock GTX 980 Ti with 337 GB/sec in actual memory bandwidth. However there are more things to consider. While the Fury X does use some form of memory compression it is not NEARLY as effective as what Maxwell accomplishes. http://techreport.com/blog/28800/how-much-video-memory-is-enough If you scroll down far enough, you will find a memory Bandwidth test, one that compares fully compressible frames with those that cannot be compressed due to how the test works, which is essentially to make a GPU render a scene where no 2 pixels are the same color, and as such are uncompressed. When compressed, the GTX 980 Ti almost ties the R9 Fury X at 379 GB/sec for the TI and 387 for the Fury X, despite the advertised 512 GB/sec for the Fury X and just 337 for the 980 Ti. And beyond that, the GTX 980 Ti can manage significant memory over clocks, with most AIB cards like the Gigabyte G1 Gaming easilly hitting 8.4 GHz in most cases, leading to a theoretical bandwidth of over 400 GB/sec, add in superior memory compression and the 980 Ti will indeed have more USABLE memory bandwidth in gaming! IF AMD had the HBM at 1 GHz it would easily be the fastest VRAM available on a consumer card even for gaming, as that would be 1 TB/sec of bandwidth! Which is EXACTLY what HBM 2 is going to reach or exceed! But at this point in time, I really wonder exactly how much difference that might make? I mean the GTX 1060 with 192 GB/sec is more than fast enough for 1440p maxed out - especially on an AIB card, or by simply raising the power target on a Founders Edition to 116%. And of course, same story for AMD with the RX 480, 256 GB/sec and it spanks the 384 on the R9 390 by a good margin, even exceeding the R9 390X in many cases. I don't doubt for a second that higher bandwidth VRAM will make GPUs perform better, but from all the data out there, it really does not seem that even with 4k gaming, beyond about 300 GB/sec makes little to no difference!
HAHAHAHA! You are great entertainment, infantile as you may be! Still responding, and you have NOTHING to say! That is what makes for the best salt in the world XD You can't say a god damned thing apart from calling me a "kid" - because nothing you say will change the CURRENT truth! And you are either too ignorant to argue the actual archtectural differences and their particular strengths and weaknesses, or you are an unfledged initiate into the brutal world of GPU competition. BTW this website? NOT A GOOD SOURCE OF ANYTHING OTHER THAN SPECS! - Though in recent times, their benchmarks have gotten better at directly comparing things - it still is irrelevant. What matters are GAMING results such as what you would have found on Gamer's Nexus which I linked you TWICE! (I think you DID read it, and that made you even more salty!) And I doubt HIGHLY you know 1/10th what I do about this sort of thing. I purposely made a few miscalculations - or to be precise planned over/under valuation of a certain figure. If you were anything APART from a puerile butt-hurt fanboy - you would have called me out. Even so, I was not far off the mark! And in the end the GTX 1060 >> RX 480 8 GB for anyone who can understand how to over clock! Thanks for playing! I look forward to more salty non-statements from you! <3
blah blah blah ......... No body cares ~(; Don't try to imply to be superior than others, because almost every single one commenting here is tech savvy. Give yourself a break! and we also need a break from you, kid! So please! :)~
31 year old actually SADDENED by AMD's failure. BTW I Just bought a GTX 980 Ti Gigabyte G1 Gaming for $330, auction finished 5 mins ago. So I honestly could NOT care less IF the RX 480 kicked the GTX 1060's ass across the board - but it DIDN'T! It fails to even hit 970 performance at 1440p or less when said GTX 970 is running at just 1450 MHz. And I never said the RX 480 was a BAD GPU, just not good enough! As for "AMD RX 480 sells well and AMD is making profit out of RX 480" - well no shit Sherlock! The RX 480 DID have a good jump on the $250 or less segment, and it DOES perform very well for the price. But ya want to know a secret? The GTX 1060 does BETTER! Similarly, a GTX 980 Ti, like the Windforce G1 Gaming I JUST bought, when both that and the GTX 1070 are over clocked to their maximum stable core frequencies (1575 on the 980 Ti, about 2055 to 2100 on the 1070) - the GTX 980 Ti is STILL a MUCH more powerful GPU! Of course I would not expect a fanboy man child with little to NO technical knowledge to understand the very strenuous practice of Over Clocking a GPU *obvious sarcasm is obvious* - so for someone like yourself, the 1070 MIGHT be a better buy than the 980 Ti. NTM the 980 Ti, gets almost IDENTICAL gains in DX12 to Pascal, because of the fact that it can process 1 Graphics and up to 64 compute commands per clock, per GPC! Which is basically identical to Pascal. Oh and MEMORY compression: Maxwell 3rd Gen makes 7 GHZ effective equate to 9.3, and you know what Polaris Memory compression manages? 8 GHZ effective to 9.5! So yeah, call me over excited, and I call you technologically challenged, and a fan boy wearing blinders that ignores REAL performance!
Are you real, dude? Is this dude even real, guys? Are you a ten year old kid that gets "overjoyed" easily with things? W/e say, kid. AMD RX 480 sells well and AMD is making profit out of RX 480. Live with it. Cheers!
What's really sad is the RX 480 only has 32 ROPs capable of just 40.5 Gpixel - at 1380 core that may go up to 46 or so. The GTX 1060 manages 72.6 at just 1506 MHz when it runs at 1709 at all times by just upping the power target, at which point it outputs 82+ Gpixel. And sure the RX 480 has better Texture output by a good deal at 182.3 versus 120.5 on the 1060 at 1506 MHz - but without the ROPS to output those "Texels" (literally textured pixels) it means NOTHING! Much like Compute performance does when it comes to game play. Basically, I forget the equation, but gaming at 2560x1440 max detail - the average game pulls between just 60 and as much as 115 Gtexel, now if you are running 4k textures that rises to about 150, which an over clocked 1060 can easily hit. A perfect example of Gtexel meaning NOTHING without enough ROPS look no farther than the GTX 770 4 GB GPU! It outputs over 144 Gtexel - but due to only having 32 ROPS, when over clocked it maxes out at just 38 Gpixel. Still a fine enough card, but it is overtaken by many it used to beat, as newer games demand faster pixel output, not having enough Pixel output is EXACTLY what causes those BAD low frame rates! That and of course a much less optimized and efficient architecture. Notice how the R9 390X is basically always behind the GTX 1060, when the R9 390X beats the HELL out of the 1060 in both Memory bandwidth AND Texture output at 184 vs abouyt 140 at stock boost on the 1060. Care to guess what it is behind in? Pixel output, and MEMORY compression! Not to forget them clock rates! And the FACT that the R9 390X can barely be nudged into an OC!
LMFAO and you are an absolute HYPOCRITE for even responding! I only READ all these comments yesterday! So many AMD fanboys talking Polaris up and it is FLOP! It is slower in almost EVERY game when compared to a STOCK GTX 1060, the few games the STOCK clocked 1060 is behind the 8 GB RX 480 on, it STILL ha better minimum frame rates! AND you responded without so much as LOOKING at the best RX 480 8GB vs the GTX 1060 there is! One that shows at stock clocks, right out of the god damned box, the WORST GTX 1060, the Founders Edition, beats the RX 480 in nearly EVERYTHING! And it gets better minimum FPS in EVERYTHING again! Go to the second to last page, where the Founder's edition is OC'ed to 2088 at ALL times, with +450 to the memory clock (giving even the FE card 220 GB/sec) - Then there is the MSI card that had a full 600 Mhz to the memory clock leading to about 235 GB per second, and BOTH cards at 2088 MHZ at all times, kick the SHIT out of the 8GB RX 480 at 1380 MHz with LIQUID COOLING - at EVERY RESOLUTION and also makes it win in basically EVERY title! I'm not even going to get into the fact that GTX 980s are selling (here in the US) for just $220! And I have a winning bid for a 980 Ti - probably the SECOND best 980 Ti ever made as it easily hits 1575 core and VRAM hits 8400 to 8800 effective, for over 410 GB/sec - at which point it actually still beats the GTX 1070 - which is faster than ANY AMD card, and basically TIED the 1080! If you cannot read Engrish, the BAR GRAPHS SHOW IT ALL! http://www.gamersnexus.net/hwreviews/2518-nvidia-gtx-1060-review-and-benchmark-vs-rx-480/page-6 GTX 1060 > 8GB RX 480 at stock speed, and 1060>>8 GB RX 480 when both are over clocked as far as possible! Oh and did you see the XFX Nitro RX 480 8GB with the 8 pin power connector, and a badass cooler?It STILL cannot clock past 1380 core AT ALL! AMD power draw is truly AWFUL! But hey what do you expect from a die shrink rebrand of an already behind GPU architecture? I mean Maxwell kicks its ass! As an over clocked GTX 970 beats the RX 480 even at 1440 in almost every title, whilst drawing LESS POWER despite being a 28nm GPU! FYI The GM204 (970, 980) and especially GM200 (Titan X/ 980 Ti) see rather LARGE performance GAINS in DX12 as they are ACTUALLY able to execute 1 Graphic and 32 compute commands per clock - and all Nvidia really needed to do, was break even between DX11 and 12 to stay in the race - but now? An overclcoked GTX 980 can keep up with, and provide MUCH more stable FPS than Fury X! So many AMD fanboys IGNORE 2 of the most important things about a GOOD GPU: 1) Stable FPS - as 45 FPS with a 35 minimum provides FAR smoother and more responsive game play than a GPU that manages 62 FPS average but drops down to as low as 12 FPS - which is EXACTLY what AMD cards, especially the Fury do! And 2) OVER CLOCKING! Oh sure it LOOKS nice to see that R9 390X beating a GTX 980 - but oddly enough ALL the major tech channels that did this review where the 390X was ahead, they not only lowered some detail settings as BadseedGaming put it "to level the playing field" - they also NEVER show Over Clocked results, because AMD cards simply CANNOT OC! So what you see is ALL YOU GET! And when Nvidia is ahead at stock? Well AMD is FAR behind overall! I WAS HOPING Polaris would be the winner, seriously! For all the crap I just talked about AMD, I WANTED to have an RX 490/490X that was both cheaper than the GTX 1070, but also drew less power and OVER CLOCKED better! I WANTED AMD to have some lead in the upper midrange! And I honestly was at one point OBSESSED with AMD thinking the ACEs (Asynchronous Compute Engines) as well as a revamped FAR more power efficient Polaris - would EASILY become king! But to my disgust to AMD, Polaris was LITTLE more than a slight improvement on the Granada architecture! Basically adding a FEW, rather unimportant features like the primitive discard accelerator, and ONE important change; Memory compression APPROACHING Maxwell's 3rd gen Delta! And it is a MASSIVE flop in the US for anyone looking for a $250 USD GAMING GPU! The RX 480 is not a BAD card at all, it just simply is NOT GOOD ENOUGH is the problem! I know in Japan Nvidia GPUs are HEAVILY over priced - the RX 480 may easily be the better deal, as again it is a DECENT GPU! But overall the performance is with Nvidia yet again, and that is ALL I am a fanboy of, that over clocking, and power efficiency - things AMD GPUs fail at thus far!
Well, with Crimson 16.7.3, I don't call less than 5% slower in DX11 average and potentially much faster in DX12 with 25% less premium is called "much slower" as you stated. And you, waited 2 months just to bite back? what a funny dude! Are you even real? :ROFLMAO:
AAAAAAND 2 months later... Not only is the GTX 1080 the FASTEST card in the WORLD - til the Titan XP then the 1080 Ti are released. Furthermore, Polaris 10, the RX 480 - AMD's FASTEST Polaris (which mostly beats the 390 and sometimes the 390X) is MUCH SLOWER than Nvidia's GTX 1060 in gaming! http://www.gamersnexus.net/hwreviews/2518-nvidia-gtx-1060-review-and-benchmark-vs-rx-480?showall=1 There are about 3 or 4 out of 15+ benchmarks where the RX 480 8GB tops the GTX 1060 Founder's Edition at totally stock speeds, even then the minimums - 1% and 0.1% low FPS (the stuttering and micro freezes) are MUCH better on the GTX 1060 than the RX 480 in EVERYTHING! Especially when one simple ups the power target setting on the FE card to 116% But read the whole way! You will find the GTX 1060 FE capable of sustaining 2088 MHz core and +450 BASE VRAM speed - which brings VRAM bandwidth from 192 to about 235 GB/sec - all this adds up to a GTX 1060 that VASTLY outperforms the RX 480 - which with the benefit of XFX nitro cooling and 8 pin power connector STILL cannot get past 1380 MHz core! All while the RX 480 TAKES MORE POWER THAN A GTX 1070! And a LOT more than the much faster but only $10 to $35 more expensive GTX 1060 ($35 more for the Zotac AMP! edition) Mind you I am speaking only of GAMING - the RX 480 does indeed have better compute performance than the GTX 1060, but unless you are bit-mining - or trying to do odd scientific stuff with a consumer grade GPU, that means nothing at all! And really, all Maxwell cards needed to beat EVERYTHING AMD HAS was DX12 drivers, which they now have! Take a REFERENCE GTX 980 (which can be bought for $210 on eBay right now) and over clock it, you WILL be able to hit at least 1450, but more likely 1500 MHz+ if you set the fan curve and power target properly, and you have a GPU that beats ALL *single* AMD GPUs even in DX12! Pascal was MUCH more than Polaris! Polaris is basically a die shrink rebrand with a few new features! As evidenced by the poor performance and HIGH power draw despite the 14nm Samsung FinFET - where Pascal managed to make 1280 CUDA and 48 ROP equal 2048 Maxwell CUDA and 64 Maxwell ROP! Mind you while the 980 may be a LITTLE faster in average FPS, it is not by much - but in 1% and 0.1% the 1060 is FAR better!
geeezus delete it hahaha
16,384 GFLOPS 5,632 GFLOPS Even though this notation is not the same as where I live, I knew the comma isn't a decimal point. https://en.wikipedia.org/wiki/Decimal_mark
... G=Giga... T=Tera
Don't count too much on the GTX 1080. Lower IPC than Maxwell by at least 10% but make up by much higher clock speed and overclock potential. If Polaris is as good as AMD claims, then its IPC is potentially very much higher than Pascal. Don't believe me? Do a simple math equation on Maxwell and Pascal. 980Ti = 2816 cores * 1075 MHz 1080 = 2560 cores * 1733 MHz 1080 / 980Ti = (2560*1733) / (2816*1075) = 1.465 ~> 1080 is supposed to be 46.5% faster than 980 Ti theoretically. And this is Maxwell to Maxwell scaling. Nvidia claims Pascal is X2 (if not X10) faster Maxwell, and yet 1080 ends up only 20~25% faster than 980 Ti. Yeah, AMD is gonna be f'ed really. Good luck with that.
Hell, no. It's more expensive than it's CrossfireX counterpart (two Nanos or Fury Xs) by about $600 in Canada and performs the exact same. Total waste of money unless you're a creator or some guy who wants to put as many GPUs as possible into a server rig
Well wait for the GTX 1080 Founders Edition and AMD is fucked again. Price-To-Performance and Performance itself.
Yeah, especially since the Pro Duo is claiming to be the "fastest gaming graphics card in the world." However, it costs TWICE as much as the 980Ti, so the price-to-performance is kind of out of the question.
anyone else super excited to see the amd pro duo benchmarks?!
LOL you guys srsly believe that GTX 980Ti has a faster memory bus? HBM at 1GHz is ALOT more than GDDR5 will ever be...
btw gpuboss its TFLOPS not GFLOPS...we are talking about GPUs not CPUs
comments powered by Disqus