Fermi - the story so far

1. High clocks, lots of transistors, revolutionary architecture - NV30 good video card make did not.
2. Nobody has ever made a great video card, late. The "comeback" is a dream yet to be realized. Radeon < GF2. fx5800 < 9700pro. X2900XT < 8800GTX, Fermi...
3. It takes more than just a year to make an ASIC, and with so many changes, odds are stacked against nVidia:

Feb 2010 launch may be optimistic.. enormous amount of testing and bugs possible with:
- GDDR5 (unlike AMD whiich has already used it for a year, this will be nVidias first)
- 40nm & 3.1B transistors & clocks (high yields are virtually impossible - praying for avg)
- lots of work required for DX11 (ie AMD had DX10.1 since HD3xx..)
- huge architectural changes to SPs to accommodate DFP performance (AMD just doubled theirs with minor tweaks)
- pressure-cooker deadlines produce glitchy product: PSU problems, mobo/chipset incompatability, XP/Vista/Win7 compatibility

Even if Fermi is perfect in every way possible and has enormous performance improvement:
- certainly, a price to match
- not even rumours of mid/low range parts - AMD already has HD57xx out last month.
- no mention of anything to counter eyeInfinity
- exclusive PhysX advantage lost since DX11 provides open standard

(just wait it can get much worse!!)
depending on extent of delay ..
- despite popular TWIMTBP game dev forced to partner with AMD since only DX11 market solution
- nVidia loses OEM contracts since no Win7+DX11 check box feature
- games are made and QA according to AMD spec - after months game dev unlikely to patch to fix nVidia bugs and nVidia likely to have more rough unoptimized glitchy driver
- nVidia only has graphics to rely on. Can't make chipsets for either AMD or Intel. No CPU business. And even last generation GTX280/260 lost bad in price/performance compared to HD4870/4850.
- super bad: last couple nVidia OEM/retail releases - parts that struggle to compete with HD4650 - already giving lame has-been impression...