My problem with you is that no matter how often I have to reiterate my points, and no matter how true and logical they are, you keep insisting on broken logic and pulling things out of context.
Either you don't get what I'm saying, and you shouldn't bother discussing with me, or you're deliberately pissing me off. In which case you also shouldn't bother discussing with me.
It seems like I'm speaking with data compression. The posts I make contain a LOT more information than what you seem to extract from them and respond to. Most of the stuff you respond to has been dealt with already, either in the post itself, or in a previous post. You're just not picking up on it.
The problem is that just because the stuff you say is logic to you it doesn't mean it is true or it is logic to everyone else.
As you see in this and other threads there is a few people that have problems with what you say.
Have you seen for example the huge amount of power the GTX 480 512 SPs requires compared the 480 SPs version in the exactly same site you cite to say NVIDIA can do GF100 with 512 SPs?
And if what I say isn't logic you just need to point to hard evidence. The problem is you don't have any as most the rest of us don't have it either. The difference is the rest of us treat what we say as rumour based, speculation and hypotheses while you treat your opinions that also have no evidence backing it up as truth.
Do you know what the yields of Cypress were at launch? Do you know what the yields of Cypress are now?
Nope. And neither do I.
Likewise, we don't know GF100 yields at launch and now. What we know is that GTX480 512 sps has been MIA.
And it is because of something related to the manufacturing process. Die size only? Well if it is die size only and if GF104 isn't that big of a departure to the GF100 (as you beleive), then the yields are related.
As a rule of thumb it is related by the following expression.
Yield_Chip_A = Yield_Chip_B^(DieSize_Chip_A/DieSize_Chip_B)
So lets say that GF104 has a yield of 80% as an example - that would mean GF100 would have a 73% yield (die size of GF100 529mm^2 and GF104 366mm^2). If on the other hand GF104 has 50% yields, GF100 should be 37% yields max. Going the other way around, if GF100 was 20% GF104 should be around 33%.
Of course as process matures, the differences from yields related to die size tend to shrink.
But again, while die size can explain some of the power consumption and yields, the fact is that NVIDIA needed more time and more respins to get a somewhat acceptable yields (if they are indeed acceptable, xbitlabs
http://www.xbitlabs.com/news/video/...Graphics_Card_with_512_Stream_Processors.html seems to imply these "new" 512 SP GF100 are actually old). It will be interesting to see if GF100 gets discontinued in favor of GF104 (which could imply that the yields problems weren't solved or simply this chip isn't a competitive product, which is less believable if the y
oelds are acceptable).
What the die size doesn't explain, if those power consumption numbers attributed to the 512 SP GF100 are true, there is something there that isn't related to dies size.
And if it isn't because die size then what causes the power consumption difference?
Facebook
Twitter