In this video: a comparison between MSAA and MLAA http://www.youtube.com/watch?v=uu1mEVrCE6Q&feature=youtu.be&a at 1:35 minutes in, you can see that even at MSAA 8X we still can see the severe aliasing of the moving objects. So my conclusion is that the sub-pixel problem in MLAA is not as bad as...
I totally agree. All games themselves are fake. It is just how you make it believable. It does not matter which route you take. Those screens look horrible by a mile ( like the first-gen games of this gen).
SSAO is not in the demo. AO is baked in but still dynamic (what's going on?)
Information from the GDC10's talk.
http://www.selfshadow.com/talks/rwc_gdc2010_v1.pdf
you can't just use elementary calculations to judge quality. But you can't deny the significant jump in quality from UC1 to UC2. That is all that matters.
GoW3 team gave you the 50% efficiency figure, that would give an estimate of improvements in quality of the next game.
Scheduling is just one of many problems they have. Imagine if they can improve most of their algorithms?
by the way, here is the definition of last minutes:
http://twitter.com/TDMoss/status/9854078121
I don't know if anyone would argue that the jump from Uncharted 1 to Uncharted 2 is just a PR talk (they also gave a conflict statement either: 30% and sometimes they said it might be misleading to give a percentage). So it is all in our head to use inductive reasoning to figure it out. I would...
You know what, at the last minutes they've got MLAA to work, and a new scheduling scheme that saved at 8% on the SPUs. If they keep improving, I don't think we are already there yet.
But what you don't understand is the original implementation has its drawbacks, whereas in GoW3 it has none (as far as I can see). Isn't GoW3's implementation better? So they must have solved it somehow right?
Remind you that in Saboteur, MLAA does not work on all edges and it's slow. You don't...
I don't think GoW3 does both MLAA and 2xAA.
What I understand from Christer is that they went beyond the proposed MLAA described in the original paper and solved the sub-pixel problem.