Safe GPU temps?

As a former semiconductor reliability professional, I'll say that every degree hotter degrades potential performance (meaning the maximum clock rates at which correct results are obtained) and lowers the probability of survival without gross failure to any hoped-for lifetime you could name.

However, in real life, lots of people run graphics cards at reported die temperatures in the low 80s for years successfully.  It does not appear that the base population (that is to say, samples without unusual manufacturing defects) has a severe cliff in reliability expectation until somewhere above that.

The more modern cards (including your and my 1060s) have internal controls which seem designed to make temperature-related severe failure unlikely, first by regulating the fans, then by altering running conditions to reduce power consumption, then by reverting to a sort of "safe mode".  I assume that the professionals designing those controls were driven by some information that had them believing they would hold down temperatures enough to avoid severe lifetime degradation to the great majority of the population.

If you are unlucky, and received a part with a much lower cliff than typical, you just might avert early failure by running at 65C rather than 75C, but for most people the only difference will be a slight difference in how much overclock you can get away with at a particular error rate on a particular work load.

I put lots of big fans in my boxes, but run them slow, so the sound does not annoy me too much.  Similarly I adjust the GPU fan controls to cool a bit more than the manufacturer settings for the lower temps and fan speeds, but avoid the banshee howl of really high fan speed.  It is summer in Albuquerque, and my study is not especially well cooled.  In the last 24 hours TThrottle reports that my 1060 has reported as high as 75C (it would do better in this box, but it shares cooling with a 1070, and both are overclocked).  I'm not very worried.