Oh cr@p, again some high tension situation. Let's tell people a bit friendlier that they might be/are probably/are certainly wrong..?
Mike, the intention is of course appreciated Some people are just really keen on correct information/facts.. That's all
I don't know the details, but I can imagine that a board is specifically designed for the specific structure of a certain chip. Tailor made for that chip design.
The right connections need to be in the right place.
If you take an Audi A6 and you feel a BWM engine would suit you better, it's really not gonna fit out of the box. That's what's being said here.
Unless you know a chip that factually has the identical physical layout AND functions. Only then you could park a different chip on there.
Poor application of cooling paste indeed causes those chips to run bad/hot.
Poor soldering connections also cause a wide range of misery. Check the first image on this page: http://www.ko-ki.co.jp/en/memo/memo_10.html
That's exactly what is suspected to be happening a lot during production if the working temperature would be too low (to respect the specs of of all other components on the board concerning max temp resistance).
The CPU may indeed run hotter then an Intel GPU, but maybe that is the case because it can perform better?
All in all, I was told there are many different ways that "reflows" are executed by 3rd parties. As it is an extremely delicate matter to find the right temperature + dureation of reflowing, to not damage components, and to really guarantee a FULL reflow of every single tin ball, you might imagine that if there are 10 people doing reflows, there might be a chance that 1 might do it really good, and 9 might do it almost good (with all the consequences of that).
A complete soldering is the strongest. Yes, a GPU that runs hot causes mechanical stress on the joints by thermal expansion. But on the other hand, ity's known by design what the temperature should be, I doubt errors are made in that design phase.
Application/production on the other hand leaves room to make mistakes, as described above. So if a perfect reflow is done (as shd have been done during production) then there is a real chance it'll last really long. Especially if you run tools like smcFanControl to draw out heat at a higher rate constantly.
I had my Mac reflown by someone who -according to what I saw and understand mechanically (I'm a mechanical engineer)- really does do 100% sure flowing of all balls, because something specific he does. I would really like to know how many of his fixes are returned within the first year.
D3us, can you say something about that?
Finally, if a good reflow would really fail after a short time, there is always the option to have the GPU removed and cleaned (or simply renewed) and reballed with "old type balls" that contain the old stuff. Then there will definitely be no more breaking of connections...