Fermi die means high Nvidia silicon costs ahead

Posted by Eva Glass in Las Vegas

There are some benchmarks. Or some Powerpoint slides. As TG Daily reported earlier, Nvidia flew in a gaggle of European journalists just when CES was all done and dusted, to brief them on Fermi, expected to be formally announced on Sunday week, January 17.

Nvidia's media strategy is pretty interesting - it didn't want to raise Fermi expectations ahead of and during CES - people might say awful things like it's delayed or something.

But industry sources close to Nvidia's plans claim to have the lowdown on the low profile briefings.

While some have said that foundries such as TSMC have let down Nvidia, the industry sources say that the low yields are not entirely due to the Taiwanese company, but quite a lot to do with the large die Fermi is yielding.

The sources claim that the cost of silicon alone will make it pretty hard for Nvidia to compete with AMD on its so-called Hemlock product, with yields something around 20 percent or so.

This would mean that Nvidia would probably lose money just on the silicon alone - it's reputed that the Powerpoint slide that journalists were shown on Sunday in Las Vegas quoted an 80 FPS on just one benchmark, with no actual demo being showed.

The same sources suggest that Hemlock will trip the light fantastic at 89 FPS - but then to see is to believe. It's all a game of smoke and mirrors.

Nvidia can't make a card based on these low yields at  $500 a card, otherwise they'd be losing money - it would have to be a $600 card. The channel reckons AMD's thing will be under $600 and that its yields are something like 70 percent.

What gives? Needless to say, neither paranoid AMD nor ultra-paranoid Nvidia will comment on unannounced benchmarks.