OK I just brushed my teeth and really am going to bed now, but I just wanna say that I hate having to write the same thing like 4 times in a row. Especially when I know you know that I got my card for non-gaming purposes; and when you act as if I recommended to OP something I didn't. I only started to comment in this thread bout the "overhype" stuff. You can have OP's specific fact pattern. I wasn't commenting on OP's fact pattern so much as the "overhype" comment. This is not about the GTX 670 which is priced $150 higher. I think we can all agree that the GTX 670 is the best card at its price point, like the 7850 is at its price point. They aren't direct competitors to each other, and we're only talking about them as if they were competitors, due to OP's wishy-washy "soft" $350 budget that apparently was actually more like $400 in the first place. Because he started off talking about the 7850 and 7870 only to stretch his budget up and say he wanted to hold onto it for 3 years later. If OP had a hard cap at $250 would we even be comparing the two cards? If OP had a hard cap of $400 would we even be comparing the two cards? No, and no. Nobody is saying the 7850 is as fast as a GTX 670. Let's be real. It's 60% more expensive, so it
better be faster!
RE: Your alleging that Anandtech is wrong and you are right: seriously, take it up with Ryan or Microsoft about the Microsoft DX11 tessellation benchmark. I have not seen anything contradicting it. In fact, within sample error there seems to be rough parity between the 7970 and 680 according to the Default setting (not Max) at HTL:
http://www.hitechlegion.com/reviews/video-cards/17774?start=8
If you are referring to various sites' usage of Unigine Heaven 3 as a proxy for tessellation testing, note that it is not a strictly tessellation benchmark and something other than tessellation can be bottlenecking it. In fact, that is what AT's tests show, since the higher-tess performance of the 7870 was unable to unseat the 7950 when it came time to run Unigine 3. That implies that the limiting factor in Unigine, at least for GCN GPUs, is not tessellation.
Similarly, providing canned benches at stock is like running Unigine 3; you don't know for sure what the real bottleneck was, whether it was geometry or something else.
And note that you keep looking at stock vs stock. Looking at OC vs OC it's not so clear; I linked to HardOCP's finding that 7970 easily beats GTX 680 in Deus Ex and ties it in BF3, both of which use tessellation, for instance. It doesn't get "realer" than HardOCP, which does not use canned benchmarks.
Now, if you want to say OC is unreliable etc. etc. then that's fine--but it should at least be noted since many of us do OC. You linked to xbit lab's OC review of the GTX 460 to show how it can in some cases outpace a 5870, and yet you don't give the 7850 or 7970 their due as monster overclockers as well? Looking at canned benchmarks at stock tells only part of the story.
As for games using tessellation in more effective ways: I'm all for it. Whether it really adds that much when you're in the game rather than examining screenshots remains to be seen, but some of those ss's do seem to use it more meaningfully than, say, Metro 2033.