Originally Posted by Buris
Charlie is well-known to hate Nvidia in the past, although some of his newer (last year) posts forecasted that Nvidia was almost a year ahead of AMD on the first to the die-shrink.
I would say there are either two things going on at Nvidia, both raise some concern.
1. They are having yield issues or TSMC is not making the chips fast enough
2. They are artificially increasing demand by keeping stock at extremely low levels, allowing the 1080 and 1070 to be sold far above MSRP
Both are possible, though I side with the latter reasoning.
I agree 2 is more likely. Finfet in general improves yields a significant amount versus earlier planar nodes, which makes up for part of the wafer cost to some extent.
If Nvidia is able to produce gp100 at this point, even if it is single digits, gp104 should not be a problem because it is vastly less complex and without the HBM stuff.
GDDR5x could be a bottleneck as most were expecting this at the end of the year for cards, not end of q2.
I am guessing, Nvidia is holding back stock to help deplete their maxwell parts. There is still way too much maxwell cards on the market and part of this is because of Polaris. Polaris 10 prehype and pricing makes it below most of the cards on clearance right now. This is making people wait, and I think Nvidia might just have to bite the bullet and release gp206 early or do an inventory write off and drop pascal prices further.
AMD on the other hand has likely stopped 28nm chips for a while because they have less invested into it at this point. They milked their rebrands for all their worth and wanted to focus on finfet. This means there is less of a write off for them at this point, and they are more willing to release new cards because there is going to be less canibalization of their cards.