There is no official announcement on this. Nowhere has Nvidia said that this was rendered on Fermi, nor in real-time (which is the key, even a very old CPU could do that given enough time).
Secondly, there are no GT300/Fermi reviews yet. No performance figures. Much of the die area is taken up with caches and interconnects that won't help gaming performance at all, it is for GPGPU. So the die will be less area efficient.
Thirdly, GT300 is yielding very poorly (about 2%, where 70-80 is reasonable). This is due to Nvidia and TSMC's problems on 40nm, all of their 40nm products are poorly yielding and late. The latest rumours say Fermi is in for a respin and won't be availible at retail in quantities above tens of units until May. Certainly not this year except for a publicity stunt.
EDIT: Update as of today: http://www.semiaccurate.com/2009/11/02/nvidia-finally-gets-fermi-a2-taped-out/
With the area handicap, and given how inefficient GT200 is (double the area of RV770 but matched in many respects by it), I would expect a Fermi to perform a little better than the 5870. But cost over $500 due to the poor yields and huge die size, and lose heavily to the dual 5870 which will be in that price range.
It's not worth waiting for, anyway.
