Quote Originally Posted by hollo View Post
could a TRU120Ex2 with double the fin depth and 12 heatpipes instead of 6 handle a phenom2 MCMed to two 4870s?
i'd say.. easily.
probably, but why would you do that?
you want to cool several chips as good as you can, so you can clock them high and max them out performance wise. then what uber genius would come up with the idea to put it all on a 40x40mm package? :P

thats what ive been saying for a long time, the whole point of FUSION doesnt add up... it makes sense for the mainstream and igp, but people always thought integrating gpus would mean better performance... well no...

both cpus and gpus are tdp limited, thats the whole point of overclocking, removing the tdp limit with better cooling and then beeing able to clock higher or bump voltages to clock even higher.

so we have two parts that are tdp limited, then how is putting them next to each other or merge them going to improve performance?

what benefits do you get from putting them on the same package or same piece of silicon?
more bandwidth...
was there a notable boost from pciE 1.1 to 2.0?
nope... then isnt it obvious that increasing the bandwidth between cpu and gpu isnt a limiting factor?

the cpu memory bandwidth is a joke compared to gpus, so again, thats limiting, not speeding things up! unless you give them seperate memory interfaces which means loads of pins and defeats the purpose of putting them on the same package. or you let the cpu use the gpu mem, but while it has massive bandiwdth the latency is terrible, and the cpu would suffer and not benefit from that. so AGAIN, it doesnt make sense...

the only good point is to save money if you put two mainstream or entry level parts on one package rather than two, or the same silicon die. and to everybody who thinks this is a revolution, amd geode gx cpus have had integrated gpus for a decade, and they are far from the only cpus with integrated gpus... its just uncommon outside of the ce segment, thats all...