GNC has less raw performance/mm^2 in comparison to VLIW architecture, that's a fact. A 28nm VLIW chip of the size of Tahiti could have packed probably 30% more ALUs.
We agree, not on the 30% though
But performance are much more consistent than before, with a much higher min. framerate, and overall I don't think that real-world scenario perf/mm^2 have declined considering the diminuished return they were getting as they increased the number of ALUs.
Indeed there is a clear increase in minimum framerate.
In the future, real perf/mm^2 for GNC will greatly surpass the old architecture, as more and more engine move to compute shader technology. Look at the performance gains AMD achieved in Civ 5, a game that uses heavily compute shader technology. It's more than 65% faster than the older architecture.
Well on the last benches of Anand review the card does indeed very well. But in the others one the difference is not that consistent (if you compare the HD6850 and the HD7770 / ~200 millions transistors including disabled SIMD and tex units, 200MHz in clock speed, or the full blown hd6870).
I do agree that it does more with less raw power.
Even if in the console space architecture are exploited in a different way, so a VLIW chip may reach more susteined and closer to its peak performance, compute is not only bound by FLOPs, but also by cache architecture, internal bandwidth and so on. And GNC (or Kepler/Fermi) offers much more on this side.
Microsoft knows where the graphics is heading. DirectX11 introduced DirectCompute, they are working on C++ AMP, an extension for high-performance computing an many other things. I'm going as far as to say that they won't have GNC in their console, but probably even a more compute orient architecture, for example Sea Island or, if they launch in 2013, something similar to that year architecture.
A fully HSA console would greatly benefit the entire AMD business.
I don't dispute that either. The chip does really great with the limitation it faces mostly the 128 bit bus vs HD 68xx. It does significantly better than HD5770 (hardware.fr gives 121 points 100 points being the hd5770 @1080p, hd6850 &70 achieve 132 and 157 points) it's 121% the perfs and 150% the transistors.
It's tough to make a fair comparison with the HD 68xx as in some games the extra bandwidth just show. the real comparison would be how perform a HD 6850 on a 128bit bus @ 1GHz, we will never know.
So the only let say fully qualified one is the HD5770 vs the HD7770 (and I go back to the first point, the 30 %). When it all said and done, no matter how promising some benchmarks are, if I take away the clock speed advantage, while no forgetting that the new gen do more how of less bandwidth and looking at the transistor count increase I'm still not sold on GCN being the way to go versus 1.5 Juniper or a tweaked Juniper.
In fact I'm also iffy about something that would be based on Barts/Kurts that could be the less "fat" of the AMD architecture (ie the Northern Island family according to anand, I remember it wrong earlier). I sounds like really good starting point too. It would have be interesting a successor to the 57xx using this architecture. It may have ended performing as well or better as the HD57xx while being tinier. In both case most likely not significantly enough for AMD to vouch it worse the effort.
Anyway now that GCN is out it's pretty hard too make a clear point (one way or the other) that it's the way to go. And from the architecture point of view the choice of something based on Northern Island doesn't sound that foolish.
In fact if want want to tweak thing s a bit, it's possibly a really good choice as there is few "fat" to the architecture, adding cache here and there over whatever depending on what you want is somehow doable. GCN is "fat" to begin with.
