Ailuros said:
Dave's answer about his own pricing estimate is quite obvious what I was aiming at.
Or put in another way - you don't know the cost differential between a 256-bit and a 128-bit bus.
That's OK, I don't know either. That's why I asked.
The question I was trying to answer for myself was if it wouldn't make a lot of sense from a product positioning perspective to, rather than supporting both 128 and 256 bit interfaces, instead support GDDR3 and GDDR4, making it supremely easy to provide two different SKUs from one chip. GPU chip, packaging and as far as I can remember board design could all remain the same, making product supply balancing painless.
Performance would probably suffer a bit vs. a 256-bit interface.
So as I was thinking about this, the economic side of the equation required some sort of handle on the incremental cost of GDDR4 vs the incremental cost of a 256 bit bus for chip, packaging and PCB. I lacked useful data. My estimates were that the cost differential was probably small and in favour of GDDR4, and definitely so over time (as GDDR4 isn't really more expensive to make than GDDR3, supply is the only issue). For all the talking that has been going on for years here, I have
never seen bus width cost quantified, but it seems to me that someone in the industry that is familiar with chip packaging and PCB cost should be able to make a good estimate. (My own could be wildly off base and is effectively useless.) I thought you might have spoken to such an insider.
