Why AA doesn't get the respect it probably deserves

[Jerry Cornelius]

This is why I lost interest in 3d gaming. Lot's of eye candy and benchmark results but less attention paid to creating a bubble that draws you into the game.

 

[Jerry Cornelius]

what it's going to take to change is for someone to publish a game that takes the gaming experience to a new level by making the graphics seemless and the game fun to play, then everyone else will follow suit.

 

[Andrew Lauritzen]

PS. here is something I googled which explains it a little more eloquently.

That link seems to largely support what I was saying, even citing temporal super-sampling as an effective way to get some motion blur and combat "temporal anti-aliasing". It also very clearly states that your retina perceives the integral of the last ~120ms of stimulus.

The discussion about motion tracking is preliminary, and somewhat tangential in any case. That's a pretty specific detail that I'd be happy to look at once we've lopped off the much more pressing aliasing issues.

 

[Simon F]

Things can move on the screen, but not on my retina.

And what are you using to watch these "things"? CRT or LCD?

Anyway, I know what you are saying but I would argue that, given N things on the screen moving in N different directions, I'd rather have N with temporal antialiasing assuming your eye was targeting the centre of the screen than N with no temporal AA at all.

 

[Blacklash2]

"Very high quality" set in up coming games ought to enable AA and some degree of AF. I'd say "High quality" = 2xAA and 4xAF, and "Very high quality" 4xAA with some degree of Super sampling and 8xAF. I think that is fairly conservative and reasonable. If you are using old cards select medium settings, or "balanced", which would not use said filtering.

 

[MfA]

And what are you using to watch these "things"? CRT or LCD?

I still have a good old CRT.

Anyway, I know what you are saying but I would argue that, given N things on the screen moving in N different directions, I'd rather have N with temporal antialiasing assuming your eye was targeting the centre of the screen than N with no temporal AA at all.

Integrating over the full time between two frames is unlikely to give optimal quality though. The integration time/weights is a completely ad-hoc and artistic decision ... there is nothing really fundamental about it. It's an effect more than anti-aliasing.

 

[Simon F]

I still have a good old CRT.

Integrating over the full time between two frames is unlikely to give optimal quality though. The integration time/weights is a completely ad-hoc and artistic decision ... there is nothing really fundamental about it. It's an effect more than anti-aliasing.

Guess we'll have to agree to differ then

 

[MfA]

If a pixel isn't a square why is a frame a block?

 

[Simon F]

If a pixel isn't a square why is a frame a block?

Who said it was? I assume you are asking how what is the shape of the filter in the time domain?

 

[Ailuros]

Those 100.87 drivers need a few manual tricks to end up 100% operational apparently, but I don't have the time for that.

I ran a couple of tests under a completely CPU bound scenario:

UT2k4, 3DCenter Primeval, 1600*1200, 16xAF:

1xAA = 62.268
4xMSAA = 62.121
8xS = 58.605
16xS = 49.949
32xS = 25.242