Do you think the NV30 will include a Wu AA-derived algorithm?

[Uttar]

Hello everyone,

After doing a lot of investigation, I found that the AA algorithm I described in another thread is nothing more than Wu Antialiasing algorithm, described at SIGGRAPH 91.

The result is good quality antialiasing at a low cost. And if you add Floating Point & Alpha Blending, Wu Antialiasing becomes a very good alternative!

In September 1992, Michael Abrash, who was a programmer for Quake and which currently works for Microsoft ( his main occupation there is the XBox ) , thinked using Wu Antialiasing to increase texture mapping quality is an interesting idea, but it could be a little too slow for the time...

trust, however, that you can see how easy it would be to
improve image quality by antialiasing with the DDA approach. For example, we
could simply average the four surrounding pixels as we did for simple,
unweighted antialiasing in this column last year. Or, we could take a Wu
antialiasing approach (see my June column) and average the two bracketing
pixels along each axis according to proximity.

Today, Wu Antialiasing united with DDA in real time is VERY easy on a CPU. And on a dedicated processor such as a GPU, it's even less costly :)

However, problems arise about alpha blending because it's order-dependant. Now, there are a LOT of possible solutions, but none are perfect. But this thread isn't about that.

This thread is about knowing if you think nVidia found such a solution, implemented it in the NV30, and will announce it Monday.

So, what do you think?



Uttar

[Postal]

For me, the order-dependant nature kills it. An antialiasing feature you can't just turn on in drivers and leave on isn't terribly worthwhile; I don't want to have to make special exceptions because a particular game gets visual glitches with a unique mode of antialiasing (isn't the "fragment" AA from Matrox contingent upon the game using it)?

[budd_wm]

nVidia is (or I should say they will say this at Comdex) claiming that AA with the GeforceFX will have no impact on performance-that is, they are claiming 0% impact on performance. Of course I dunno how they will do this, but I do know that's what they said (they haven't publically said it yet, but they will). I wonder if 3Dfx had any part to play in this....

[Dr_Colossus]

I don't think it'll matter, the antialiasing might end up being programable which means just about any type of AA is possible (I think I remember a cineFX white paper hinting at such a feature). The coolest feature would be if powerusers could program it to suit there taste.

[StealthHawk]

Originally posted by budd_wm
nVidia is (or I should say they will say this at Comdex) claiming that AA with the GeforceFX will have no impact on performance-that is, they are claiming 0% impact on performance. Of course I dunno how they will do this, but I do know that's what they said (they haven't publically said it yet, but they will). I wonder if 3Dfx had any part to play in this....

how do you know this already :confused:

but as i keep saying, we'll know in less than 48 hours. all bets will be taken. and those that need to eat crow will have it stuffed down their throats forcefully ;)

[Uttar]

Postal: Of course, it cannot be used as-is. That's why Z3 or a strange type of buffer would fix that. I'm just saying there are OTHER ways than Z3 to fix that.

budd_wm: David Kirk said that was their goal with the NV30 about a year ago. I don't remember him saying 0% hit penality, I always remembered he said something more like 10%... But that IIRC, I could be wrong...

Dr_Colossus: That's partly why Wu Antialiasing would be so darn great. You could change a LOT of values.
Here are a few examples:
1. Minimum influence to be written on screen ( the smaller this number, the lower is performance )
2. Minimum influence to recalculate texture information ( this would hit fillrate, but increase quality )
3. Quality of associated buffers

And maybe a few other things.


Also, if we wanted nearly free AA, we'd need concurrency. The most logical way this can be done would be to do AA during Pixel Shading. How can you do that?
Well, I've got a little idea... But that seems very unlikely.
It would be a bug-fixed FAA, with a dedicated pipeline to work on all antialiased pixels.
So, the penality would, amusingly, be to fillrate. Hmm, but I still couldn't call that "free"... Unless you call -10% for excellent quality AA free...
In other words, I really don't think this is what nVidia will announce. But then... what?


Uttar

[budd_wm]

Well the info I recieved is what they will announce at Comdex. I don't have all the details though, but I do know it will have way better performance than the 9700 does with AA. Most of what the Inquirer has said is true, but they got a few things wrong.