VirtualLarry
No Lifer
- Aug 25, 2001
- 56,554
- 10,171
- 126
I'm not a chip designer, but as I understand it, they give the fab a "design blueprint" for the chip, called a netlist or something, yes.Originally posted by: Slaimus
NVIDIA only presents the design of the chip.
Then the netlist or whatever, is processed by the fab, and adapted to whatever process tech that they are using - I think - to eventually get the new set of masks used to create the chips.Originally posted by: Slaimus
It is up to each fab to decide how the actual chip wil tap out. Small differences in transister spacing can lead to the extra length. They could have even cut the wafer differently.
But the point is, those mask-creation steps and whatnot are expensive, so generally, if one is going to rev the "design blueprint" of a chip, unless it's a serious defect (in which case earlier batches of the chip may have to be scrapped altogether), one waits until there is some other change at the fab that would necessitate another mask creation. (I think.)
Well, that was the question - was that id from the older or the newer GPU? The intriguing point is, that -if- NV was going to fix the NV40's PVP with a revision, then a fab change would be the *perfect* time to do so.Originally posted by: Slaimus
There are also many suppliers of silicon wafers, so the difference could be just that. If they fixed the VPU, it would have a new revision. Seeing how it is still A1, I doubt that happened.