Theres almost smeg all difference between a "gpu" and a "gpgpu", and all the talk suggesting otherwise is rather silly.
If you want to do tesselation in an accerated piece of hardware, or in software on non dedicated hardware, is just a choice, they weren't identical architectures before(AMD/Nvidia) and they aren't now having their own choices being made.
We don't know for sure how much less of a "gpu" it is yet anyway. It may well have everything in dx9-11 hardware accelerated, it might not, it might not and still do them very well in software, or it might be utterly crippled doing it in software and suck big time in newer games. I would suggest something in the middle, a competitive card that takes a bigger hit than the 58XX through to 3 generation old parts of ATI's when using tesselation and other things, it probably won't suck though.
The end result isn't the interesting factor or the important one though, as a company, the viability or selling millions of cards at cost to maintain sales, well, isn't very good long term. Fermi, slow or fast, doesn't look like it can realistically turn a profit and thats really what Charlie et all are talking about. Charlie hasn't bashed it saying its slow, it probably won't be, but he can't see how a 50% larger core with far more complex PCB can be sold competitively priced against AMD parts. is 5-25% more speed, worth 25-50% more in cost?
Will a 395gtx with two cores blow away a 5870x2, probably, will it be required, no, will the 5870x2 be able to be sold at a higher profit than a 380GTX? These are the important questions when thinking about the "future" of Nvidia as a gaming company, and they are valid questions that have worrying answers.
How many people on this forum would pay £400 for a Fermi, when a 5870 offers 80% of the performance and is dropped to £200-250?
Its not just is Fermi better than the 5870, its how much better at what cost.
Maybe Fermi can beat it and Nvidia screw AMD on pricing by selling it at £200 aswell, but how long can Nvidia do that?