You're basically reposting the same thing you said above. Spam?The performance of these mov.(GS)
Shader Performance RadeonHD3450 vs GeForce8500GT vs Chrome540GTX
http://www.youtube.com/watch?v=S8cUp-8Fyr0
Radeon HD 4670 vs GeForce GT240 vs Chrome540GTX Shader Performance
http://www.youtube.com/watch?v=ZNHoi8iEHrk
and
L1,L2 which is connected directly with shader in the picture(separately from Texture Cache).
GF100 has greatly improved geometry performance. I'm not sure how that data point leads to the conclusion of a read/write cache in Chrome. I suspect you're reading too much into a marketing diagram because the specs would surely specify read/write cache if it exists.As for Fermi, a Geometry performance largely improves by Cache.
There is GSCubemap as the one example.
http://pc.watch.impress.co.jp/img/pcw/docs/343/002/ph15.jpg
And there is 540GTX which shows a good result in GSCubemap in mov.