Digital Foundry isn't respectable?
They just have that little fcat graph on screen but they don't do any frametime graphs, 99 percentile numbers etc. And they rarely offer any insight. Without detailed frametime measurements benchmarks are pointless.
Here's an example: gta 5 runs at about 60 fps average on the radeon pro duo at 1440p ultra. (a 2 gpu on one PCB graphics card), which is about 50 percent faster than it runs on a fury x.
Sounds good, right? And that's all the info most benchmark sites will ever give you, along with maybe minimum fps.
But if you check pcper's benchmarks you see this:
HOLY FUCKIN SHIT, it's the worst stuttering you have ever seen in any game on any gpu, ever. (the green line, no... green wall)
The pink line is how it runs on sli gtx 980 tis for comparison
This is why you need frametime graphs, because it shows that the 99 percentile frametime for this radeon pro duo is much lower than that of the fury x
http://www.pcper.com/files/review/2016-04-29/GTAV_2560x1440_PER.png
The fury x will give you an infinitely much better experience
If there is anything wrong with the 1080's drivers or frametime performance you'll have to go to pcper or techreport to find out. What if for some reason the frametimes suck compared to maxwell, and you currently had a 980, would you want a 1080 ? HELL no because who cares about higher average fps if the 99 percentile maximum frametime is worse.
Similarly maybe pascal will be able to offer more consistent frametimes than maxwell, or somehow have less cpu overhead or something , which would make it more attractive.
DF stumbled upon an i7 absolutely killing it frametime consistency wise compared to an i5 in this video
https://www.youtube.com/watch?v=mSU0UWFCa1Y and they didn't even notice, too busy talking about the i3.