Apart from the higher cost, there is a benefit to running a constrained video card:
1) higher end video cards have more shaders (these make up the processors) on the order of many. GPUs can have well over 50 cores. Because of this, you have a bigger processor.
2) larger size means more surface area to radiate heat.
3) comparing the 1080 vs 1060 as an example:
2560shaders vs 1280
3dmark11 28458 vs 14843
TGP 165 vs 80
Rest of the parameters still the same
From overclocking & underclocking experience, you need a lot of power to top the performance, but based on this same logic, you can cut a lot of power without having to underclock much. You can reduce the power down to 80w, but only need to drop the speed about 25%. Given double double the cores, you take about 15-20% performance reduction. Now these numbers are no where near accurate, but it gets the point across.
My past experience involved dropping the power on my GPU by 25% while only dropping about 15 % performance. The first 10% required a fast downclock, but after that, I could really crank down the voltage without reducing the GPU & meme speed much at all. Eventually you reach a point where the card won't balance the power use correctly & you have to add a 10% factor of safety over that point to keep it stable...I ended about 25% down. This is how manufacturers can can play the developers game in reverse & still sell a decent computer.
