Our buddy Fudo over at The Inquirer reports that ATI and NVIDIA will be taking dramatically different approaches for their next-generation flagship GPU's. On one hand, we see ATI lobbying developers to focus more on the number of Pixel Operations instead of the number of Pixels per Clock. In contrast, NVIDIA is pushing for the exact opposite approach to be favored by games with their heavy emphasis on the number of Pixels per Clock. Regardless of which design approach is used, I think it is safe to say that we'll be seeing some insanely fast GPU's showing up the first part of the year.
The G70, Geforce 7800 GTX has 24 texture memory units and 16 ROPs and can theoretically draw eight pixels with each, having three textures per clock. G71 will be able to do even more pixels per clock - it's as simple as that.
ATI's R580, on the other hand, will still be able to push sixteen raw pixels per clock but will be able to render 48 pixel operations per clock. So ATI will insist on more per pixels operations before you actually draw the pixel. This kind of makes sense when you need more maps and calculations per pixels.
I would fear the very successful and expensive "The way it's meant to be played" program which Nvidia spent some $160+ million dollars on last year. With that kind of budget it's much easier to convince developers and publishers that Nvidia's marchitecture is the best. We can surely say that 2006 will be just as full of marchitecture wars as 2005 was.