The 6 series definitely has better scaling... those suggesting its better than SLI tho need to take a closer look - if you exclude the oddities both come out very similiar on average gains - which is a big step up from the 5 series granted that always lagged behind. However its not the whole story... SLI is still producing better minimum fps in most cases where it counts and has a more consistant average fps without being pulled up by silly high scaling in low complexity scenes.
That said its very close and cayman based GPUs might even pull ahead of SLI with a bit more work.
http://www.anandtech.com/show/3987/...renewing-competition-in-the-midrange-market/8
what you said needs even further explaining though, the 460gtx has better min fps than the 6870, but then you realise they didn't include the stock one but the over 20% faster overclocked 460gtx in there.
Either way a single card gives unplayable minimums, again this irks me about Anandtech, why no xfire/xli results at the highest res, why, because they are morons.
Ok, down to the 1920x1200 res, the 6850 has higher than 460gtx 1gb minimums in Crysis, and in xfire, the minimum framerate goes up just OVER 100%, the 460gtx 1gb minimums go up UNDER 100%.
So that suggests to me the 6850 has both higher minimums, and the minimums scale better in xfire than SLI.
I'm fairly sure there will be a game two you can find the reverse, but when you look at the overall scaling, and the fact that in both reviews the 6870/6850 scaling beats 460gtx 1gb scaling in almost every single game, I'd imagine that the minimums will stay rather similar aswell.
The minimum frame rates are also, its hard to be certain but the 4 way shader of the Cayman is likely to increase minimum throughput by 20%, infact thats potentially one of the biggest area's of increase in Cayman, if its already beating Nvidia in comparable cards, and gets a 20% bump, well, we really have to see how that works out in the end.
As for the suggesting of scaling increase as new cards are released, thats perfectly valid, AMD obviously did something very specific to increase scaling beyond driver differences(you assume, unless they are filtered down to 5xxx in the future).
The problem being, the 580GTX despite Gibbo's protestations, seems to be nothing more than a base layer respin of fundamentally the same architecture. The 6870 is a pretty medium, but important step from the Cypress architecture, 35% more efficient is pretty huge on the same process. Nvidia doesn't seem like it will be sporting any significant changes from the previous architecture, which means the likelyhood of new core logic increasing scaling is, well, not very likely.
Theres also the IMHO, bigger issue, of diminishing returns, that is, the closer you get to 100% scaling, the smaller the increases you will get. AMD had a lot more room for improvement, again the 5870 scaling is actually very impressive in some games, less so in others, in the worst games it had 20% to improve, in some games its gone from 75-95%, AMD won't really get hugely better scaling because, well, theres only 5% to go and outside of extreme synthetic benchmarks where you can get 100% scaling, 95% is basically as good as it gets. Both SLI/Xfire are generally speaking as good as you could really ask for right now, the only issue being a few games that really doesn't seem to suit it, but then things like CivX firstly don't need a high framerate and should be(not played it, stab in the dark because of it being RTS) utterly cpu limited.
If both companies in generall get a 90% boost from the second card, or higher, well a 2nd card gives you very good value, 95% isn't really going to change anyones decision, neither is, 95 up to 97%.
But then we have the future, a significant change to architecture can bring around unforseen bottlenecks, which with them could bring serious multigpu scaling troubles which would then need fixing again. But right now, if they manage to keep it in check, sli/xfire is almost "finished" averaging 90-95% is all you can ask for, all you need and realistically all you're going to get.
EDIT:- I wasn't really looking at the other results but, the 470gtx has horrible/the worst minimum's scaling, the 5850 is only marginally worse than the 460gtx 1gb, the 5870 somehow has worse minimums in single card and actually scales the best out of all of them, and the 6870 is just over 100% aswell.
Either the balance of shaders vs bandwidth is marginally different and they just got a slightly weird result for the single card minimum, or I don't know, they'll normally run the bench 3 times and take an average. I can't really explain that one, even if you add a couple fps to the single card minimum to bring it inline with where you'd expect it to be, its would still have very good scaling. However, the minimums in both the 6850/6870 favour better and are closer/above the 5870/5850 when you compare overall speed where they are clearly behind their namesake cards. Minimums are better, max's are a worse and average is a little behind with the new gen cards.
