Antialiasing : How it works on modern cards?
 

I remember from voodoo5 days that AA filtering was made possible by drawing
an image at high resolution, say 4 times the normal display res. and then
scaling it down, thus eliminating the jaggies and getting smooth picture.
This however was a very slow process.

Can someone tell me is it the same process in current DX9 cards, like 9600,
9700, 9800 and 5700 5900?
Also does radeon8500 uses the same methos?

Actually, the Voodoo5 used RGSS/JGSS in it's FSAA implementation. Oh..
RGSS=Rotated Grid Super Sampling, JGSS=Jittered Grid Super Sampling. Rather
than render the image at 4 times the resolution (upsample) to the offscreen
buffer and scale it back down (downsample) while averaging the subsample
data around each pixel (this is OGSS or Ordered Grid Super Sampling), RGSS
takes subsample pixel data at points of random rotation around the pixel.

There are religious wars even still over which implementation is better from
a tech standpoint and from a practicality standpoint... RGSS or OGSS.... but
no one will dispute the superior IQ afforded by RGSS.

Anyway... 3dfx used RGSS. All other manufacturers (ATI and Nvidia) used, and
AFAIK still do use, OGSS.

Correct me if I'm wrong, but I'd assume all ATI cards are doing OGSS.

--
Tony DiMarzio



"Asestar" a s e s t a r @ s t a r t . n o wrote in message
...
I remember from voodoo5 days that AA filtering was made possible by
drawing
an image at high resolution, say 4 times the normal display res. and then
scaling it down, thus eliminating the jaggies and getting smooth picture.
This however was a very slow process.

Can someone tell me is it the same process in current DX9 cards, like
9600,
9700, 9800 and 5700 5900?
Also does radeon8500 uses the same methos?

Thanks for reply. That does explain it.
However I tought that with this new breed of gpu's everyone is talking about
supperior performance with AA enabled, compaired to previous gen gpu's.

Example: a r8500/gf4ti4200 would perform somewhat same as r9600/Fx5600 in
certain cases, when AA is not enabled. However enbling AA makes newer cards
so much faster than the previous generation.

This led me to believe that there are some revolutionary change in AA
techinique with new DX9 cards. Maybe they use some kind of pixelshader
programming or whatever to do AA? Guess not then.


"Tony DiMarzio" wrote in message
...
Actually, the Voodoo5 used RGSS/JGSS in it's FSAA implementation. Oh..
RGSS=Rotated Grid Super Sampling, JGSS=Jittered Grid Super Sampling.
Rather
than render the image at 4 times the resolution (upsample) to the
offscreen
buffer and scale it back down (downsample) while averaging the subsample
data around each pixel (this is OGSS or Ordered Grid Super Sampling), RGSS
takes subsample pixel data at points of random rotation around the pixel.

There are religious wars even still over which implementation is better
from
a tech standpoint and from a practicality standpoint... RGSS or OGSS....
but
no one will dispute the superior IQ afforded by RGSS.

Anyway... 3dfx used RGSS. All other manufacturers (ATI and Nvidia) used,
and
AFAIK still do use, OGSS.

Correct me if I'm wrong, but I'd assume all ATI cards are doing OGSS.

--
Tony DiMarzio



"Asestar" a s e s t a r @ s t a r t . n o wrote in message
...
I remember from voodoo5 days that AA filtering was made possible by
drawing
an image at high resolution, say 4 times the normal display res. and
then
scaling it down, thus eliminating the jaggies and getting smooth
picture.
This however was a very slow process.

Can someone tell me is it the same process in current DX9 cards, like
9600,
9700, 9800 and 5700 5900?
Also does radeon8500 uses the same methos?

Well.. I'm sure there not doing AA with shaders :)

I don't know the specifics of how Nvidia and ATI have optimized their FSAA
OGSS implementations. I don't think anything revolutionary has occured. But!
it does seem like there is some magic going on somewhere doesn't it? The
performance hits just aren't what you'd expect.

--
Tony DiMarzio


"Asestar" a s e s t a r @ s t a r t . n o wrote in message
...
Thanks for reply. That does explain it.
However I tought that with this new breed of gpu's everyone is talking
about
supperior performance with AA enabled, compaired to previous gen gpu's.

Example: a r8500/gf4ti4200 would perform somewhat same as r9600/Fx5600 in
certain cases, when AA is not enabled. However enbling AA makes newer
cards
so much faster than the previous generation.

This led me to believe that there are some revolutionary change in AA
techinique with new DX9 cards. Maybe they use some kind of pixelshader
programming or whatever to do AA? Guess not then.


"Tony DiMarzio" wrote in message
...
Actually, the Voodoo5 used RGSS/JGSS in it's FSAA implementation. Oh..
RGSS=Rotated Grid Super Sampling, JGSS=Jittered Grid Super Sampling.
Rather
than render the image at 4 times the resolution (upsample) to the
offscreen
buffer and scale it back down (downsample) while averaging the subsample
data around each pixel (this is OGSS or Ordered Grid Super Sampling),
RGSS
takes subsample pixel data at points of random rotation around the
pixel.

There are religious wars even still over which implementation is better
from
a tech standpoint and from a practicality standpoint... RGSS or OGSS....
but
no one will dispute the superior IQ afforded by RGSS.

Anyway... 3dfx used RGSS. All other manufacturers (ATI and Nvidia) used,
and
AFAIK still do use, OGSS.

Correct me if I'm wrong, but I'd assume all ATI cards are doing OGSS.

--
Tony DiMarzio



"Asestar" a s e s t a r @ s t a r t . n o wrote in message
...
I remember from voodoo5 days that AA filtering was made possible by
drawing
an image at high resolution, say 4 times the normal display res. and
then
scaling it down, thus eliminating the jaggies and getting smooth
picture.
This however was a very slow process.

Can someone tell me is it the same process in current DX9 cards, like
9600,
9700, 9800 and 5700 5900?
Also does radeon8500 uses the same methos?

it does seem like there is some magic going on somewhere doesn't it? The
performance hits just aren't what you'd expect.

Exactly the reason for my first post! I would love to get some more info
about this.. but from where? The old 3dfx site used to be so helpful back
then .. sob

http://www.beyond3d.com/forum/viewto...&view=previous

this should shed "some" light.

--
Tony DiMarzio



"Asestar" a s e s t a r @ s t a r t . n o wrote in message
...
it does seem like there is some magic going on somewhere doesn't it? The
performance hits just aren't what you'd expect.

Exactly the reason for my first post! I would love to get some more info
about this.. but from where? The old 3dfx site used to be so helpful back
then .. sob