Originally Posted by kyleb
it's a different demo Rollo, different demos give different benches. also, all the partial precision was put into the arb2 path so the nv30 path would just be redundant.
I agree on the different demo part.
I don't remember reading about the partial precision being put into the path nV3Xs now run though, just the nV10 and 20? Do you have links?
I remember Carmack saying he dropped the nV30 path and went to ARB2:
but not that he incorporated the nV30 features into ARB2? I'm not a programmer, but I would think this would change the definition of ARB2 and make it vendor specific again?
What happened with the 9800Pro now being as fast as the 5900U would seem to support that the 5900U is running ARB2, which it can't run as fast as nV30 path?
Please correct me if I'm wrong, don't want to be unenlightened.