I was just wondering about something. Given the specs of the average current generation cpu's and graphic cards( about 2,4GHz cpu's and the 9000 series Radeons, with 64-256mb), at what point do normalmaps start getting more efficient than polygons. I mean, if you have a model, at what number of polygons, it'd be more efficient to add detail via normal maps?
What would run better: a 20.000 polygon model without per pixel lighting or a 4000 polygon model with normal maps and per pixel lighting?

I ask this because, I don't know much about shader programming and as such don't always have the ability to combine parts of shaders, to get the desired effect. Because of that it could be that I can't use normal mapping on a model, because I want a fixed function effect on there as well. Ofcourse, the two effects could be merged into one shader, but I can't do it.