Alright so I'll try to do some quick math. First let us assume that the amount of cores/shaders has an exact direct (linear) relationship to framerate. We'll look strictly at DX12 games and look 1440p resolutions for these games b/c there was some released RX 4XX naming scheme decoded thing that came out and the RX 480 is supposed to target 1440p gaming. These benches will be from TechPowerUp for Rise of the Tomb Raider and Anandtech for Ashes of the Singularity. I'll just post pictures here:
Alright so we'll set the baseline for performance expectations based on linear scaling of CUDA core count for Pascal. According to the leaked slides, the GTX 1060 will be a half die Pascal chip (1280 CUDA cores) while the GTX 1070 is 3/4 of a die (1920 CUDA cores) and the GTX 1080 is full die (2560 CUDA cores). Not sure why I included the 1080 count but whatever.
Anyways so ideally we should expect the GTX 1060 to have 2/3 the performance of the GTX 1070 (1280/1920 = 0.6666...). I'm too lazy to check the scaling from the 1070 to the 1080 so again we'll assume the linear model.
Also we will ONLY compare the 1060 to the 8GB variant of the RX 480 because the GTX 1060 will supposedly use 8Gbps VRAM and the 8GB reference variant of the 480 does the same (4GB variants use 7Gbps memory on the reference models). I don't think 6GB vs 8GB will make too much of a difference at 1440p (take that with a grain of salt).
Alright onto the napkin math. So for the GTX 1060 with linear shader to FPS scaling we can expect 42.666666...fps at 1440p at whatever graphics quality TPU is running (we'll just say 42.7 for simplicity's sake) so basically neck to neck with the RX 480. In Ashes of the Singularity at 1440p in Extreme Quality (DX12) we can expect 35.6 fps which is quite a bit lower than the RX 480 and right on par with the GTX 970.
Alright so around the same performance maybe depending on what game in DX12 but overall the GTX 1060 wins here because of the supposed much higher efficiency of Pascal. Price will be another story but we won't know till release.
The issue right now is that there's only a handful of DX12 games right now so the immediate value of the RX 480 is kind of moot until the next couple years of game releases (which is honestly very little time so it's quite future proof). As for the current DX11 games the RX 480 kind of gets stomped by Pascal.
For example, TPU bench for Crysis 3 at 1440p because I think we all know the line "But can it run Crysis?"
Alright, once again assuming linear shader to performance scaling and ignoring the possible impact of 2GB of VRAM, GTX 1060 will supposedly get 28.266...fps (28.3 fps) so about a 10% difference if we use the calculated 1060 fps number as the denominator.
Onto Witcher 3 at 1440p.
So 40.5fps expected from the 1060 and about a 7% performance difference.
So the difference here looks a bit small with DX12 performance between the 480 and 1060 being about the same and the DX11 performance favoring NVIDIA a little bit.
Next we move to power consumption. Too lazy to link TPU's RX 480 review image of the power consumption but GTX 1070 and 1080 idle at 6W, RX 480 at 15W. Power consumption while under the labelled "typical gaming" scenario has the 1070 at 145W, 1080 and 166W, and the RX 480 at 163W.
Alright so we're looking at a pretty power hungry card here for the performance considering it gets stomped by the 1080 using the same amount of power as well at the 1070 which uses much less power. Maybe the 1060 will draw something like 110-120W or less.
So in the end that ends up scaling as far as overclocking goes. Nowadays it's pretty easy to overclock since overclocking software often has set profiles and there's a large enough community to give you a good average of what you can expect most cards will be able to clock up to w/o any issues. Given that Pascal is much more efficient for each watt you make the card draw you'll end up with more performance clock per clock added on Pascal over Polaris. Not only that but since the 1060 draws less power it possibly has quite a bit more headroom.
Even worse we look at the stock coolers. The Polaris stock cooler looks nice but is quite lousy in performance not only being somewhat meh in cooling potential but also quite loud. The 900 and 1000 series stock coolers are quite well made so immediately the RX 480 has very little headroom using the stock cooler compared to NVIDIA's cards. Of course aftermarkets are going to be available but I still feel like the 1060 will fare better simply because it's using a more efficient architecture as far as FPS per watt goes.
tl;dr So in the end while performance may end up about the same and more in NVIDIA's favor in DX11 scenarios, that extra OC headroom from efficiency for the 1060 makes it pull ahead.
Just speculation and napkin math based on a linear scaling assumption that probably isn't correct so again, take this with some salt. Missing a bunch of other factors like the 2GB memory deficit the 1060 has under the 480 8GB.