Ok, should have read the complete thing before posting, they don't use any 3dfx patents, it was more "Voodoo like patents" but still close enough to mention. Also after looking at some of the bench marks, it's not quite up there but who knows, I'm going to keep looking for more info on this one. Still might be worth checking out.

About their slowness..

Remember when the 8500 came out?
It's drivers were so poor, that it performed somewhat near the GF3 TI500.
Nowdays it can easily beat up a GF4TI4400

well, haven't read the whole article yet, but concerning these voodoo-like patents:
the vsa-100 2-way/4-way renders per scanline, i.e. the chips work together on the same picture (suppose evryone here knows that   )
the v8 duo is said to work another way, in fact, one chip render exactly one picture and after giving it to the ramdac, renders the 3^rd one (due the second chip has rendered the 2^nd picture already, and so on...so each chip must process harder than any scanline interleave)...to my mind 3dfx's method is more effective & i suppose xgi would have used it if they would have got the rights for.

yeah, i know, 'cause that's exactly what i said..
anyway, yours is really theoratical, thus not containing the practical use --> think hereof:

Quote:

wrong in practise! let's say, one picture, rendered by the 1^st chip shows a detailed image of an explosion. that could take its time to be rendered. so the second chip has to wait...sure, it can render the 2^nd picture already, but can't send it to the ramdac before the 1^st chip has send his picture. another thing is, sometimes you got framedrops (frames already rendered but not of use anymore as the required scene has changed too fast).
that doesn't matter with scanline procedures: both chips work together to put the 1^st picture on the screen and then work together on the 2^nd one. you see, 2 chips render a picture faster the only one chip, but 2 chips rendering each another picture brings nearly no advantages..thought 3dfx's method brings 166% while xgi's must deliver about 125-133% of one chip's power
yet, i'm sure ati & xgi would like to use scanlines but quantum3d won't be so stupid 

Hmm...

I 'think' SLI does have a slight advantage over AFR in terms of efficiency, but I'd have to test it to be sure.

If I had the Time, I would do a Scalability test with a Voodoo2-SLI, Voodoo5 and an ATI Rage Fury MAXX.
Unfortunately, time has become a luxury, despite all the Hardware being readily available to me

What about SLI'ed triple buffering How would you put THAT? 

Quote:


A hint: the triple buffering does exactly like "2) Box-per-worker (Alternative frame) " between 2nd and 3rd buffers.  The 1st buffer is only for display. Now apply SLI to this scheme 

So I guess 3dfx made the right choice with SLI, cos is subect to further improvements (too bad they never made triple buffering a reality... although it was planned)