AA names help

[darckhart]

@taltamir:

afaik, SLI AA just means the load is split between the cards (much like AFR) for AA (ie, SLI 8xAA = 4xAA on each card). the caveat is when SLI AA is enabled, sli frame rendering is disabled. (yea weird i know.)

afaik, nvidia's naming goes like so:
16x = 4xMSAA + 12xCSAA, 16xQ = 8xMSAA + 8xCSAA
32x = 8xMSAA + 24xCSAA, 32xQ = 16xMSAA + 16xCSAA
64x = 16xMSAA + 48xCSAA, 64xQ = 32xMSAA + 32xCSAA

 

[sticks435]

That was my impression too after doing all my research. The only one that doesn't follow this convention is 8x, which is 4+4, but that could be because 8x msaa already existed.

 

[bryanW1995]

Great thread, sorry I missed it when it came out. Keep up the good work Tal!

 

[busydude]

http://www.tomshardware.com/reviews/anti-aliasing-nvidia-geforce-amd-radeon,2868-4.html

Tom's take on CSAA and EQAA.

 

[sticks435]

http://www.tomshardware.com/reviews/anti-aliasing-nvidia-geforce-amd-radeon,2868-4.html

Tom's take on CSAA and EQAA.

Well that freaking sucks All the time I spent the other night trying to figure this out just got wasted.

 

[bryanW1995]

Didn't Don Woligrosky work at Anandtech in the past? I was going to throw out the typical "Tom's has gone downhill" statement before reading that article, but it was actually quite good.

edit: n/m, looks like he writes their "best graphics card for the money" articles. I must have just recognized his name from that.

 

[Skurge]

Can anyone tell me why I have the option for EQAA with on my 5770. I thought only the 6900 series cards supported it.

 

[(sic)Klown12]

Can anyone tell me why I have the option for EQAA with on my 5770. I thought only the 6900 series cards supported it.

You can edit the .inf in the driver package or change a registry setting to have it show up in CCC on any 5xxx or 6xxx card. RadeonPro might also do the same thing when it ask you if you want to enable unsupported features. If you select it, you'll only receive normal MSAA

 

[Idontcare]

http://www.tomshardware.com/reviews/anti-aliasing-nvidia-geforce-amd-radeon,2868-4.html

Tom's take on CSAA and EQAA.

Is it just me or do they all look like ass?

 

[BFG10K]

Can anyone tell me why I have the option for EQAA with on my 5770. I thought only the 6900 series cards supported it.

They do, your 5770's ROPs physically don’t support the sampling of such a data type.

 

[taltamir]

added FXAA.

 

[rootz91]

thanks for the knowledge dude
but a budget card owner like me won't have time to meddle with those :'(

 

[Pottuvoi]

There seems to be quite big error on description of MSAA.

There is only one difference SSAA and MSAA.
SSAA calculates shaders/textures on each subsample of a pixel.
MSAA calculates shaders/textures once per pixel and copies the result to all subsamples. (shading location is middle of pixel.)

MSAA is only visible on edges, because that is the only place where subsamples can have different values.

Z/Stencil is done per subsample in both MSAA & SSAA and thus both handle AA correctly even when polygons are rendered in random order.


Also small errors on FSAA.
FSAA was old description for full screen antialiasing which just describes all techniques which work like SSAA or MSAA or similar.
Doing downscale from 2x vertical & horizontal resolution has 4 subsamples per pixel and thus would be 4xAA. (actually a 4xSSAA)


T-Buffer AA / Accumlation buffering AA, an old good accumlation buffer method which renders image multiple times with slightly jittered vertex positions.
These images were additionally copied to single buffer and then result was divided with amount of images copied to the buffer.
Result is SSAA with free sample locations which you can scale from 1 to n samples/pixel without additional memory consumption outside one deep buffer.

This tech is what is now used on games like Gran Tourismo and others to create photomode pictures.

 

[taltamir]

There seems to be quite big error on description of MSAA.

There is only one difference SSAA and MSAA.
SSAA calculates shaders/textures on each subsample of a pixel.
MSAA calculates shaders/textures once per pixel and copies the result to all subsamples. (shading location is middle of pixel.)

MSAA is only visible on edges, because that is the only place where subsamples can have different values.

Z/Stencil is done per subsample in both MSAA & SSAA and thus both handle AA correctly even when polygons are rendered in random order.

I am going to need more then just your word for it. I based the data I posted on sources.

Also small errors on FSAA.
FSAA was old description for full screen antialiasing which just describes all techniques which work like SSAA or MSAA or similar.
Doing downscale from 2x vertical & horizontal resolution has 4 subsamples per pixel and thus would be 4xAA. (actually a 4xSSAA)

I did not make a mistake, I listed different CONFLICTING Sources and what each says about it. For me to make a mistake I would have to have misparaphrased one of them.
I get that you have your own view on what is the correct explanation, but I am sourcing things. Gimme a source and I will add its claim.

 

[Pottuvoi]

I am going to need more then just your word for it. I based the data I posted on sources.

Here you go.

Wiki about MSAA.

The specification dictates that the renderer evaluate the fragment program once per pixel, and only "truly" supersample the depth and stencil values.

Wiki about SSAA.

Samples are taken at several instances inside the pixel (not just at the center as default) and an average color value is calculated.

Microsoft about MSAA.
nice paper about basics of SSAA and MSAA.
Very old one.
From opengl forums.
opengl.org.

In case of hardware MSAA.
MSAA reduces shading cost by using same result on all subsamples within a polygon when writing into a color buffer
Combined Z/Stencil buffer (24bit/8bit) is supersampled to get proper proper occlusion information. (This is not necessary on other rendering methods which can occlude without need of Zbuffer, but it is how current hardware works.)

I did not make a mistake, I listed different CONFLICTING Sources and what each says about it. For me to make a mistake I would have to have misparaphrased one of them.
I get that you have your own view on what is the correct explanation, but I am sourcing things. Gimme a source and I will add its claim.

Previous sources should clean this up as well.

In general when talking about over sampling AA methods the amount of subsamples is told with the 1x, 2x, 4x, 128x, just like amount of subsamples when doing anisotropic filtering and such.

Additional very nice link about MSAA, SSAA, HDR.. etc.
http://mynameismjp.wordpress.com/2012/10/24/msaa-overview/

 

[tweakboy]

Great OP thread! good stuff you dug up!

Ok Im using FXAA ,,,, enabled in nvpanel and have CSAA 8x

When I have fxaa on, what AA I pick.. right now I have it to 8x CSAA and FXAA on , multisampling.

thanks guys Im still confused over this CSAA no CSAAA or FXAA no or yes etc... thanks for helpers.. gl

 

[taltamir]

Pottuvoi, I need primary sources not a wikipedia page that is rather short of citations.

I will read the good links you posted and add their info later on.

 

[darckhart]

entries needed for TXAA (no idea what abbrev means) and SMAA (subpixel morhphological AA).

here's the nvidia page for txaa, but no real tech description:
http://www.geforce.com/landing-page/txaa

here's the page for smaa:
http://www.iryoku.com/smaa/