Downsampling or super-sampling down from a higher resolution is a brute-force form of anti-aliasing. Nvidia's new DLDSR aims to do the same job with lower pixel counts, improved performance and maybe improved quality. What does it do? How does it work? Can a downsampled 1620p really look as good as a downsampled 2160p? And what happens if you combine DLSS with DLDSR? Alex Battaglia has answers!
- DSR from 2014 is ordered grid super-sampling AA
- 60% performance loss on a 2060 when rendering 2160p
- non-whole number scaling can cause issues because of uneven division
- DLDSR targets the filtering phase of DSR to correct errors by uneven scaling
- there seems to be edge detection and edge anti-aliasing
- DLDSR is 3% slower (on a 2060) than DSR at the same settings
- 0% smoothing does create errors
- smoothing filter is recommended
- DLDSR has temporal stability over DSR
- competitive to DSR 4x
- some resolution-dependent aspects can't match the higher internal res of DSR 4x, however; depth of field is improved with higher internal res, thin lines exhibit less holes due to more pixels being shaded
- using DLDSR with DLSS can improve IQ while minimizing performance loss
- DLDSR only offers 1.75x and 2.25x scaling, while DSR has a wider range of scaling factors
Last edited: