By using this site, you agree to our Privacy Policy and our Terms of Use. Close

Forums - Gaming Discussion - Understanding Anti-Aliasing

Scoobes on 14 March 2013
Kynes said:
Scoobes said:
Sooo... anyone know why FXAA causes texture blurring?


Because it's a blur filter. A clever implementation of a blur filter, but it does not calculate more geometry/color samples, as other AA methods do. The good thing about FXAA and other shader AA methods is that they have a very low performance hit, and they also help with specular/shader aliasing, but they fail miserably with subpixel geometry and you can have texture blurring. There is an evolution of FXAA called SMAA that provides better image quality: http://www.iryoku.com/smaa/

Thanks for the info. SMAA has more of a performance penalty though doesn't it?




Get your Portable ID!

Around the Network
Kynes on 14 March 2013
Scoobes said:
Kynes said:
Scoobes said:
Sooo... anyone know why FXAA causes texture blurring?


Because it's a blur filter. A clever implementation of a blur filter, but it does not calculate more geometry/color samples, as other AA methods do. The good thing about FXAA and other shader AA methods is that they have a very low performance hit, and they also help with specular/shader aliasing, but they fail miserably with subpixel geometry and you can have texture blurring. There is an evolution of FXAA called SMAA that provides better image quality: http://www.iryoku.com/smaa/

Thanks for the info. SMAA has more of a performance penalty though doesn't it?


SMAA has 3 modes: 1x, 2Tx and 4x. SMAA 1x should be almost as fast as FXAA, on a low performance graphics card FXAA penalty should be less than 5%, SMAA 1x ~7%. SMAA 2Tx has a bit higher performance hit, ~10%. SMAA 4x has a rather big performance hit, between 2x MSAA and 4x MSAA, but with a much higher image quality IMO.

Take a look at this Crysis 3 review, where they compare the different AA options in image quality and performance: http://www.hardocp.com/article/2013/03/12/crysis_3_video_card_performance_iq_review/



Scoobes on 14 March 2013
Kynes said:
Scoobes said:
Kynes said:
Scoobes said:
Sooo... anyone know why FXAA causes texture blurring?


Because it's a blur filter. A clever implementation of a blur filter, but it does not calculate more geometry/color samples, as other AA methods do. The good thing about FXAA and other shader AA methods is that they have a very low performance hit, and they also help with specular/shader aliasing, but they fail miserably with subpixel geometry and you can have texture blurring. There is an evolution of FXAA called SMAA that provides better image quality: http://www.iryoku.com/smaa/

Thanks for the info. SMAA has more of a performance penalty though doesn't it?


SMAA has 3 modes: 1x, 2Tx and 4x. SMAA 1x should be almost as fast as FXAA, on a low performance graphics card FXAA penalty should be less than 5%, SMAA 1x ~7%. SMAA 2Tx has a bit higher performance hit, ~10%. SMAA 4x has a rather big performance hit, between 2x MSAA and 4x MSAA, but with a much higher image quality IMO.

Take a look at this Crysis 3 review, where they compare the different AA options in image quality and performance: http://www.hardocp.com/article/2013/03/12/crysis_3_video_card_performance_iq_review/

Very useful info, thanks! Looks like SMAA is the future. I've been using it at 2Tx on Crysis 3 but had no real clue how it worked or how much of a performance hit my system was suffering. From those shots though, even 1x looks good!




Get your Portable ID!

TheJimbo1234 on 14 March 2013

And here I was expecting you to talk about Fourier transforms. Ah well, maybe next time.



Kynes on 14 March 2013
TheJimbo1234 said:
And here I was expecting you to talk about Fourier transforms. Ah well, maybe next time.


But we're talking anti aliasing on videogame graphics, not anti aliasing on a mathematical way. I don't think that this forum is the best place to talk about the Shannon theorem or signal reconstruction.



Around the Network
TheJimbo1234 on 14 March 2013
Kynes said:
TheJimbo1234 said:
And here I was expecting you to talk about Fourier transforms. Ah well, maybe next time.


But we're talking anti aliasing on videogame graphics, not anti aliasing on a mathematical way. I don't think that this forum is the best place to talk about the Shannon theorem or signal reconstruction.


..it's done using the same method. How do you think they code it? "Please blur this pixel? Cheers buddy!" =p  



haxxiy on 14 March 2013
TheJimbo1234 said:
And here I was expecting you to talk about Fourier transforms. Ah well, maybe next time.

Yeah, because talking about integrals and transforms will surely make up for much better conversations here on a videogame forum, right?

People just need to know the core of anti-aliasing programming so they don't go around spreading bullshit and misleading others, that's all. 



 

 

 

 

 

Kynes on 14 March 2013
TheJimbo1234 said:
Kynes said:
TheJimbo1234 said:
And here I was expecting you to talk about Fourier transforms. Ah well, maybe next time.


But we're talking anti aliasing on videogame graphics, not anti aliasing on a mathematical way. I don't think that this forum is the best place to talk about the Shannon theorem or signal reconstruction.


..it's done using the same method. How do you think they code it? "Please blur this pixel? Cheers buddy!" =p  


I know, I know, but as haxxiy says, I'm sure that 99% of the people here doesn't know what a FFT is, so it's not the way to try to explain it. "intelligent blur" AA isn't really AA, it's not a mathematically correct way to do AA, but it's a good enough substitute. Real AA as MSAA or SSAA do it the "correct" way, taking more samples to try to correct the undersampling.



RazorDragon on 14 March 2013
Weedlab said:
Brings me back to PS2 days. Oh so jaggy.


Not if you played on a CRT screen.



TheJimbo1234 on 14 March 2013
haxxiy said:
TheJimbo1234 said:
And here I was expecting you to talk about Fourier transforms. Ah well, maybe next time.

Yeah, because talking about integrals and transforms will surely make up for much better conversations here on a videogame forum, right?

People just need to know the core of anti-aliasing programming so they don't go around spreading bullshit and misleading others, that's all. 


Very true, I just thought it was obvious, hence the presumption he would go into how it is done and the mathematical variants of MSAA and standard etc and why the performance demands rapidly increase.