Originally posted by: BFG10K
At this stage 400 MHz really means nothing in terms of trying to estimate how much faster it'll be than the 9700 Pro.
Yep, that's right. It'll have 8-pixel pipelines, though, and have 2 texture units per pixel pipeline and have more than sufficient memory bandwidth.
That flash fillrate means nothing if the card will be starved for memory bandwidth. The NV30 will have real problems if the memory is only 128 bits wide compared to the 9700's 256 bit memory interface unless nVidia has something special up their sleeve to get around this issue.
I'm expecting nVidia to use some of Samsung's 800MHz DDR or perhaps a bit faster, to push theoretical memory bandwidth up to around 13GB/s. It'll still be a good 7GB/s or so lower then what the R9700 is capable of, but I figure nVidia will try to use some of GigaPixel's technologyto boost bandwidth efficiency to somewhat off-set the advantage ATi will have in sheer theoretical bandwidth.
Even now a good couple years after, some of GigaPixel's patents are quite impressive and could really benefit the NV30.
I'm not sure it'll be quite enough to fully match up to the R9700's real owrld memory bandwidth, but I don't think it'll fall -too- far below.
I don't expect it to be too badly memory bandwdith limited though, except in extreme cases.
Of course this is all speculation on my part...
The second TMU should of course help out to considcerably varying degrees depending on the scene being rendered.. if sufficient bandwidth is there of course.