It's the FX episode all over again.

Not Canadian anymore

I've also seen rumoured prices for $499 for the 470 and $679 for the 480. I think these are closer to the mark personally.

Anyone can come up with numbers at this point. For example, my "sources" say they will charge $1 and $2 for the 470 and 480. Expect this to be the top story on all major tech websites tomorrow.

--

It's funny how the OP's article is now seen as a beacon of hope for Nvidia. A few months ago everyone would have thought 5870-level performance was far too low for the second-highest card in Nvidia's lineup.

A few weeks ago people would have also said that the HD 5830 using more power than the 5850 and almost as much as the 5870 would be insane but a few days ago we found out 1120 Stream processors, 800Mhz core clock, 16 rops and 175W board power. However Charlie said so, maybe hes right. In any case it does fall in line with how bloody hard to produce and expensive the Fermi chip is.

It si not a question of making up numbers. Inside rumours have been floating around for months now that TSMC's yields are anywhere from 5% to 20%, mostly cornered at the lower end of the percentage scale. It is also rumoured that NVIdia uses "risk wafers" for its first batches of chips.  Given that there are only around 100 chips on a $5000-6000 wafer, every workable chip costs NVidia anywhere from $250 to $600. I doubt NVidia is going to sell the stuff for $19.99 ...

I'm aware of the rumours and the economics. Nvidia will lose money on each card whatever the price, because it won't be faster than a 5970 even with extreme optimism so they have an upper bound on price there.

But with an extremely limited run (<8000 cards being the rumour) they can sell at any price they want; if they wanted to do a PR stunt (and we've seen far too many of those from Nvidia) they could undercut the 5870 on RRP and then never have any stock in to actually sell.

Yields are chip-specific anyway so you should really say Fermi's yields rather than TSMC's. e.g. I'm certain Cedar is yielding above 80% despite being 40nm.

The truth is probably somewhere in the middle which in this case is still a failure for Nvidia. For now it seems that Ati is the way to go.

The 5-20% yield is in connection with the Fermi chip. It is rumoured that about 7 to 10 chips are fully functional per waver and another 10-20 chips could be functional with one cluster deactivated. The yield very likely isn't even close to 80% for dual-cluster-deactivated chips.

Yeah, I just mean you shouldn't trust anyone who says TSMC's yields are X (like Bright Side of News). Only in connection with a die.

I said Cedar was >80%, that's the smallest Evergreen die, not Fermi. Fermi will be a lot lower in any case because it is a bigger die.