I guess people have forgotten that gtx 1060 6gb is over 10% faster than the gtx 970.
and the 970 itself is quite a bit faster than the gtx 780
with that in mind we should expect the RTX 2060 to at least outperform the gtx 1070
That's what we'd all want, but we're not getting that without a significant price hike. The 970 was Maxwell 2, which was a major leap in efficiency (and thus how much performance you get in a given power budget, due to clock speed increases) over Kepler, even if it was made on the same node. Pascal was another major leap going from 28nm to 16nm, fitting more cores in the same area and increasing clocks yet again. Turing moves to 12nm, which is essentially just a refinement to 16nm, with (as we've seen in reviews of the 2070 and above) zero efficiency improvements over the last generation, no real area gains, and no clock speed increase (>5% top OC). In other words, they need more, lower clocked cores or hotter cards with better coolers to beat the previous generation - and the die will be bigger (and thus more expensive) no matter what.
GTX 780: GK110, 551mm^2, 2880 cores cut to 2304, ~900MHz, 288 GB/s memory
GTX 970: GM204, 398mm^2, 2048 cores cut to 1664 (28% decrease), ~1.3GHz (44% increase), 196 GB/s memory (32% decrease)
GTX 1060: GP106, 200mm^2, 1280 cores (23% decrease), ~1,5GHz (15% increase, though with GPU boost closer to 1.8 in reality, so a 38% increase), 192GB/s memory (2% decrease)
So: both of the last generations have used arch or process improvements to match or beat previous higher-tier performance through clock speed increases while reducing core counts to lower costs. This isn't happening this time around, so the die needs to grow. This used to mean that Nvidia ate some margins for a generation while they waited for a new node shrink (think GTX 700 series), but that's not happening this time - they've grown too greedy and callous for that, which the RTX series has shown clearly.
For the high end core counts went from 2816 cores (980 Ti) to 3584 cores (1080 Ti, up 27%) to 4352 cores (2080 Ti, up 21%). The story for the lower tiers is roughly the same, with 1664 -> 1920 (^15%) -> 2304 (^20%) cores for the '70s. That tells us the 2060 is likely to get around 1280 * 1,2 =1536 or * 1,3 = 1664 Cuda cores, which is nowhere near enough to beat the 1070 (with 1920 Cuda cores) given the tiny improvements in clock speed and core performance from Pascal to Turing. A 1920-core 2060 would likely be too close to the 2070 for comfort - or at least it would force the 2050 Ti to suddenly become a midrange rather than entry-level card.