It has the same components as SMAA T2x, but handles TAA much better, so its better.
I am still not convinced SMAA 1tx is better overall. They said in the PDF it's a more robust temporal solution (not better or more effective AA solution) and that it's similar to TXAA temporal AA.
Still why the 1tx instead of the 2tx? Normally the number is linked or at least an indication of the sampling or theorical sampling of the AA.
On the SMAA reference PDF, in addition to the usual 2 frame buffers (back and depth) present on all solutions, SMAA 2tx needs
4 additionnal render targets (compared to 2 render targets for SMAA 1x and 0 for FXAA).
On the recent Crytek PDF they say about SMAA 1tx
accumulate multiple frames in an accumulation buffer
so they just need one additionnal 1080p buffer and "merge" each times the new frame in this unique buffer (there is weighting involved, it's more complicated than that of course), that's probably why TXAA is a bit blurry.
In summary: SMAA 2tx needs 4 additionnal render targets to store previous sampling (2 for the regular morphological SMAA +
2 1080p previous frames for TAA?, not sure here), hence the 2tx when SMAA 1tx would need 3 additionnal render targets, 2 regular SMAA +
1 temporal 1080p frame, hence the 1tx.
- SMAA 1tx: better, more stable temporal reprojection but slightly blurry all the time (like TXAA)
- SMAA 2tx: sharper, more effective AA solution but with occasionnal temporal AA artifacts.
I know which one I would claim as better
but I agree that some people may prefer the softer look of TXAA.
