Regarding the whole CPU/GPU temperature discussion (chrisnyc75 and JackNaylor), I think a good and simpler analogy is liquid flow and bottlenecking.
Consider a big container like a bottle or something, that you can fill from the top and it has a hole for liquid to escape at the bottom. After a point, no matter how much liquid is poured in, the flow of the liquid will almost exclusively be dictated by the size of the hole at the bottom. When the liquid is poured in fast enough that it exceeds the capacity of the bottom hole to let it escape, the container will start getting filled, despite the fact that there is a way for the liquid to escape.
Similarly with heat transfer. The GPUs spread their heat on a much larger surface (>400mm^2 for high-end GPUs and <150 or even 100mm^2 for most CPUs) so it's easier for it to escape towards the medium one puts over the die for cooling. There are other factors in play, but they account for negligible amounts of heat, so just having that in mind is enough.
One must also not forget that when the cooler itself is hot (i.e. when touching the heatsink of a CPU cooler and it's warm or hot) its efficiency drops, sometimes rather dramatically. The bigger the deltaT between the die and the cooler is, the more efficient the cooler will be at moving heat from the die to environment.
Surface area of the cooling apparatus, be it a heatsink or water cooling radiators/blocks is a huge factor because of the point about efficiency. When there is a huge area available for heat dissipation, heat is more evenly spread out and is way easier to be moved towards the environment, thus resulting larger temperature differences between the (hot) dies and the (cold-ideally ambient temperature) radiator. However, that only applies until the cooling is saturated with heat which again, is much harder with large surface area water cooling setups. This can easily be verified if someone with a wc loop just turns off their fans and see how long it takes for the system to overheat. I bet it will last a while under even full load before problems from overheating start manifesting. That is because, given enough cooling surface area, even passive heat dissipation is huge compared to air cooling desings.
Finally, regarding current GPU air-cooling solutions... They're basically crap. But they can't get much better without becoming huge and heavy. Fact is that if one put a CPU heatsink on a GPU they'd get amazing temps (I tried this once with an old NVidia 6800 Ultra whose cooler was broken - VRMs didn't like that solution though), but if they put a GPU cooler on a CPU somehow, the result would be less than pleasing I guess. Still, these can't be directly compared because they have quite different designs and purposes, but you get the idea.
Edited by tolis626 - 3/28/14 at 4:25am