name='Sihastru' said:
Dual GPU will not be possible, if we take into account the published TDP for the Fermi based Teslas. TMSC needs a little time to work a bit of their magic into the 40nm process before dual GPU for Fermi will be possible.
The usual trend, since '08, is to base the dual gpu on the gpu under the very top one. Laced into an interface, pcb, that is good enough for the top card. However, Fermi is a new architecture to which none of this could apply, as we don't know.
As to TSMC, all the news ever published within the tech community is based on nothing, and has been fedback by the tsmc representatives as nonsense. (mostly sourced from TI & SA or SI) In the financial sector, they have presented nothing but good news, news about the 23nm process and the massive cash they're making.
name='Sihastru' said:
Also they might not need it. If the single GPU top card will perform close to the 5970 (I know it has a small reserve of clockspeed), the dual-GPU is going to be a hard sell. Considering all developers flee towards consoles, and no games will require so much GPU processing anymore, ofcourse.
It's arguable that the majority of gaming needs nothing more than a 4g GTX285. But dev moves on, card manufacturers, or 1 of them atleast, have been working with these devs which will bring us a number of dx11 releases. Know how poorly games are both written and optimized, I can see why they want to go faster and faster.
The dual gpus are meant to offer that much more than gaming also. Details of these kinda things are also still rumors and promises.
The Fermi is based on an architecture that, to this point, doesn't exist. It can't be compared to any existing gpu situation other than end result. GPU power in terms of gaming, probably needs to get faster to compensate for poor writing or game-engines.
I would say this tho, the emphasis in 2010 is probably more involved around quality as opposed to playing CoD @ 150 fps compared to 130 fps - which has been a nonsense for a few years now.