SMAA is light years above FXAA.
Not at all. FXAA , SMAA , MLAA are all post process AA methods that functionally work more or less the same : edge detection on the final image + blur filters. Differences in performance and quality are negligible , they are all just as "cheap" really. But my claim is right in absolute terms , SMAA is more expensive than FXAA just not by a whole lot in that sense you're right.
SMAA is simply less aggressive than FXAA , gives less blur because of more refined detection (probably why you think it looks better) but fails at smoothing most of the rough edges since the amount of sample data is the same , just one 2D image . I have never seen any game where SMAA is "light years" ahead of FXAA , if you have tell me which and I'll take a look.
Basically it goes like this :
SMAA - slightly higher cost , less blur , less effective
FXAA - lower cost , more effective , more blur
If you think SMAA is better , by all means do so , but that's simply subjective. They both fail and succeed.
There is a good reason why SMAA ends up being avoided on consoles. Given that a good chunk of games run at sub 1080p resolutions which exacerbates aliasing , your best bet is to blur the whole image as much as possible hence FXAA is used.
Only way to make SMAA look good is to give it a temporal pass as well or combine it with low levels of MSAA . But by itself , it's just as bad as FXAA most of the time.