What is MSAA?

[Tatchan]

Pls, can someone point out what's the difference between MSAA and SSAA, and if xbitlabs's article on FSAA state everything correctly? Thanks

 

[David G.]

MS=multi sampling and SS=super sampling .

Xbit are usualy a serious site .

 

[Tatchan]

Yes, I know the full name of it, and SS is to render a larger framebuffer then filter it down. How about MSAA?

 

[Kristof]

Very briefly :

With Super Sampling you do a new full texture lookup for each subpixel, with Multi Sampling you re-use the same texture lookup for all the subpixels.

So with SS you can get, for example, 4 slightly different shades of red for all the subpixels, with MS you would get the same shade of red for all 4 sub-samples.

K~

 

[Tatchan]

Very briefly :

With Super Sampling you do a new full texture lookup for each subpixel, with Multi Sampling you re-use the same texture lookup for all the subpixels.

So with SS you can get, for example, 4 slightly different shades of red for all the subpixels, with MS you would get the same shade of red for all 4 sub-samples.

K~

So, MS could result in worse IQ?

 

[Hellbinder]

ahhhh... This is From Tomshardware's preview of the 9700

Besides the special sample patterns and the Z-buffer sampling, Smoothvision 2.0 comes with another special feature that ensures superior FSAA on Radeon 9700. A patented gamma-correction algorithm ensures that the color gradients generated by the FSAA procedure are displayed as smoothly as they are supposed to be.

Not really addressing the specific issue of the same red color.. but this should deal with the problem to some degree???

What do you'all make of the Patented Color gradient algorithm?

 

[3dcgi]

The gamma correction trick sounds like a logical thing to do although I'm curious if 90% of the population could see the difference.

 

[Simon F]

]ahhhh... This is From Tomshardware's preview of the 9700

Besides the special sample patterns and the Z-buffer sampling, Smoothvision 2.0 comes with another special feature that ensures superior FSAA on Radeon 9700. A patented gamma-correction algorithm ensures that the color gradients generated by the FSAA procedure are displayed as smoothly as they are supposed to be.

Not really addressing the specific issue of the same red color.. but this should deal with the problem to some degree???

What do you'all make of the Patented Color gradient algorithm?

Assuming I've understood what's being said here, how can they patent gamma correction with downsampling? It's obvious!
I was doing that in my last job (back in 1990-1) where I had to make thumbnails/icons of selected frames from a video. I converted (with some approximations for speed) from the original image to linear colour space, downsampled, and then converted back to gamma space. I think applications like Photoshop also do this kind of thing (storing the image in linear space internally and converting back on saves).

They are either doing something exceptionally clever or the USPTO is, well... enough said. To be fair though, I shouldn't comment without actually reading the patent.

 

[KimB]

So, MS could result in worse IQ?

The basic result is that SSAA improves the visual quality of the entire scene.

MSAA only improves the visual quality of polygon edges, and does not affect textures in any way.

MSAA is usually combined with anisotropic filtering to make up the difference. MSAA + aniso can generally provide a better visual quality/performance ratio than plain SSAA, or even SSAA + aniso.

 

[jb]

MSAA also has issues with Alpha textues as it does no AA on them. When the software companies use a different method (alpha blend) this is not an issue however there is a very slight blur to those texures then. Every from of AA has trade offs.

 

[KimB]

Alpha blends still look better, even when there is no FSAA enabled. The additional blur is no more than is normally introduced with bilinear filtering.

In particular, alpha blends look vastly better when the picture is viewed at a distance/high angles.

 

[jb]

Of course that is just your opinion. I prefer them with a AA instead of blending. But that is my opinion. To each their own

 

[KimB]

The blurring is only visible if the texture pixels are much larger than the screen pixels.

In other words, if you put your face into a fence, it's going to look blurry.

But so will any wall that doesn't have very high-detail textures.

Besides, it takes special programming to implement alpha blends anyway (not much, but some). You'd think that developers would pay attention and use higher-detail textures for alpha blends than they do today with alpha tests (texure detail is on the rise, anyway, so it's not like it's a huge difference).

And even using simple alpha blends on lower-detail textures look quite a bit better than none. Have you seen the alpha blends I enabled in UT? They look far better in motion than alpha tests.