It's just that mastrdrver's assumption that Fermi "will really decimate the 5 series in heavy tessellation" is a bit ridiculous at this point.
After rereading the Tech Report Fermi article I was like to revisit this for a second.
Charlie's big thing was the RV800 has a dedicated tessellator. In fact, ATI clusters the whole setup engine before the entire SMID array (adding in an extra rastorizer). Though, the extra raster doesn't increase triangle output.
Contrast that to Fermi where each of the 16 shading clusters has a tessellator. Then, it has 4 rastorizers so max theoretical triangles output is 4 per clock cycle.
So, what happens when we start getting carried away with DX11 and the ability to radically increase poly counts? Fermi has 16 tessellators to extrapolate the simple poly map into XXX amount more. Fermi can also raster just short of 4 polys per clock.
Scott (Tech Report) sums up the whole thing nicely.
Obviously, the GF100 is a major architectural transition for Nvidia, which helps explain its rather difficult birth. The advances it promises in both GPU computing and geometry processing capabilities are pretty radical and could be well worth the pain Nvidia is now enduring, when all is said and done. The company has tackled problems in this generation of technology that its competition will have to address eventually.
If that great feature of DX11, the ability to increase poly count exponentially from a simple map as its the only thing everyone is talking about, then your going to need to greatly increase geometry and poly throughput per clock. Which means your going to have to make it highly parallel. Now, how you go about doing that is up to them. One thing is certain, unless ATI can make that setup engine extrapolate more than what it has right now, we are going to be back at RV670 all over again in DX11 titles with tessellation. It will get even worse when you take away SMID arrays to make up the lower end cards. Having a monster poly and geometry ability on the front end doesn't help at all when you take away half of the SMID so you can make the lower end parts. You end up with a highly bottlenecked part that is chocked.
Why do all the rumors keep saying the 6 series is an architectural change for ATI? Cause they even realize it. Like I said they don't have to do it the same way as nVidia, as I would doubt they would. Though, geometry and poly throughput is going to need to be able to increase while not suffering shader performance. It will be interesting to see how the R900 does compared to Fermi (once nVidia gets it running right :shadedshu)