PDA

View Full Version : Speculation on nV30 and problems/performance


Spiritwalker
03-01-03, 12:04 AM
I had this thought while posting at R3d and figured that I would post it here for some tossing about:


Originally posted by Spiritwalker at Rage3d
If you look at the transistor count it is plausible that the nV30 was actually designed as a 256 bit part. There has to be a use for all those transitors. I truely think that there is a lot more borked in the nV30 than is being let on and the psuedo 8x1/4x2 is functioning well off where it should be as a result.

kyleb
03-01-03, 12:39 AM
i am prety sure that they planed 128bit all along and were just taken off guard by ati. in my openion the transistor count is most likely the result of extended ps/vs units. granted it is all guesswork until nvidia desides to come clean on things.

Bigus Dickus
03-01-03, 01:33 AM
My guess is the extra transistors are there because nV expended a tremendous amount of effort and resources making the NV30's pipelines more flexible and "general" than the previous fixed function pipelines of the past. Actually, I think nV had probably intended for the pipelines to be used in arbitrary combinations to work optimally in a variety of situations (i.e., an 8x1, 4x2, 2x4, 4x1 + 1x4, 2x2 + 4x1... etc.). I think they probably failed, and though the pipelines are very flexible in theory don't function nearly as independently in practice as was hoped, probably due to a design flaw or design choice that was made or left broken due to time constraints.