New pic of X800 Pro ( 2nd )

"Damn it man, I need more! We must overtake the R420!

Divert all power to the GPU except for life support and the PR department..."

 

LOL, classic.

Anybody like to take a stab at how benchmarking is going to be handled if Temporal AA requires v-sync and turns itself off under a certain framerate? :?

...otherwise amusing nV 8x vs ATi 8xT comparisons will be impossible.

 

Pfft! Easy-cheesy. Just hot-wire your tricorder to be a temporal field stabilizer and/or temporal phase inverter and you're good to go! 8)

 

Well, if that's the case Dave should send "the ensign that we've never seen on any epsiode before" to do the first benchmarking...

 

nV alternated AA every frame already--Xbit or Digit-Life documented that in a GF4MX article, IIRC. It may well have been before April 2003. They only did it with 4xSSAA, though, alternating between 1x2 and 2x1 SSAA, and according to the authors the difference between frames was very visible. So I'd imagine you'd need a very high minimum framerate and probably a higher sample pattern than nV's aborted attempt.

 

Well, that's not really surprising. They say that stochastic antialiasing requires at the VERY least 8 samples, more like 16, and you need a decent framerate, too.

 

I'd have thought the required frame rate would be dependent on the contrast between the pixels on alternating frames.

With a reasonable number of samples (say 4 or 8 ) there's probably an image quality win at 60Hz. I guess NV could probably do something similar even if they don't support programmable sample positions (I have no Idea if they do) just by changing the output pixel center say by 1/8 pixel on alternating frames, although that would probably have an adverse effect on text.

 

Sure, and I believe the article I was talking about noticed the "artifacts" on dark mountains in front of a light sky. The point is, you never know what the contrast will be between pixels, so you have to aim for a certain (high) level if you want to maintain the illusion. I guess 2x1 and 1x2 at 10x7 ain't high enough.

 

Would it be possible to somehow save the AA samples from frame to frame in a buffer and blend the previous two samples for the final output? I would assume some kind of rolling average would be required to cut down on the flickering. I don't know about such things but sounds like the data could be combined in a shader of some sort.

 

Thinking about this, there would be absolutely no benefit to a moving portions of the image using this system, since in any given frame you still have the same number of AA samples, and any movement will likely remove any perception of the alternating pattern.

Sure you end up with a sort of cheap motion blur effect, used in PS2 games among others, doesn't really help the spatial AA much.

 

I would think the quality would be some function of how much the frame changed each time. Higher framerates should get rid of that smeary blurry slo-mo effect at lower speeds.

 

8)

 

Voodoo 5, IIRC?

I can't believe nV hasn't done something with T-buffer yet, or at least thrown in V5-type SSAA as an awesome bonus for older games. This is assuming V5 AA is far better than 4xS, and thus worth the effort....

 

Just what I was thinking! This sounds much like the T-buffer from 3DFX, and given that their staff went to both NV and ATI shouldnt be surprising.

One question thought, Ive always wondered, the V5 AA was pure hardware, and the one from all of today cards is software? Thats wy I cant use it in Diablo 2 for example?
If thats the case, why hasnt any company implemented it as in the old V5?

And what about rampage`s m-buffer??? nah... I miss 3dfx

ps: btw, hello!