This will help me when calculating how much each new feature addition or upgrades to specs, the extra average 35% the 9700 carries per clock over the Ti4600! At least we can eliminate 8X from it, which makes more sense:
-A good 30% comes from better occlusion and LMA II, so it now has the same performance per-clock than Ti4600 (remember Radeon 8500 has a big GPU and mem clock in contrast to the GF3.)
-Now from here we start digging up the 35%:
-10% at least from the extra 2 vertex shaders.
-15% from the double bandwidth
-Worthy to note that the 128MB (proven these days that it ain't gonna provide THIS much except for new texture hog games and AA but not in general gaming with no AA or Aniso) adds a good 5%.
-5% more from the 16 textures per pass, though it can go higher since it depends whether the app uses massive multitexturing. I suppose in Aquanox, the card will rule more .
I forgot the other features, but we can clearly see where it's all coming from. At first I was wondering why it isn't having more from 4 vertex shaders having FP capabilities and 1024 instruction-width. When I now think of it, these are much more for image quality rather than performance enhancements, but in basic, the extra 2 provide a good boost.
--
Intel and AMD sitting under a tree, P-R-O-C-E-S-S-I-N-G!