Originally Posted by ricercar
I did't expect to double my fps. What's up? Can someone explain it in simple language, please? Guess I didn't pay enough attention to the old threads claiming that the NV30 has more raw _____ (¿fillrate?) than the NV35/38, despite the 128 bit memory bus, and now I want to know ....
The NV36 stands to the NV30/35 like the GeForce2MX stood to the Geforce2GTS
The original NV30 performs up to 8 Z/stencil operations simultaneously, where the NV36 only four. The 128bit bus is not a limiting factor during Z/stencil writes. But there is more: NVidia claims that the NV36 can render up to 4 pixels per clock cycle like the NV30/35 but only two of the pipelines appears to have loopback capabilities: in all the situations where multiple passes are needed (i.e trilinear filtering, multitextured surfaces) this card operates in a 2x2 configuration. This is also true for pixel shading ops, since only two of the pipelines appear to have shaders ALUs.
So, when you compare a fully fledged 4x2 card like the NV30 to a highly optimized 2x2 card you should see quite a difference in performance.
I suggest to try this fillratebenchmark on both card to see the difference: http://www.geocities.com/ndawinteractive/