Improved superscale (DR comes nearly last in reviews)

For example in this one. I thought it was okayish, better than nothing, but Topaz gives much nicer results. Can you work on the algorithm?
https://www.provideocoalition.com/a-i-u ... -shootout/

It's a shame that test didn't specify (as far as I could tell) which settings were used as it's not just on/off in Resolve. So I think we can assume default settings were used. I tend to set both Sharpness and the Noise Reduction setting to "Low" instead of the default "Medium".

I'd also be interested to know if the output is better with tensor cores than without, or is it just faster?

Yeah probably default. I think there is a lack of sharpness in added details on medium.. but low works better because the little added sharpness is also a bit unnatural.

Isn't Topaz like, 10x slower?

Is this maybe a "Be careful what you wish for" scenario?

(So far I've been impressed with the results of 2x SuperScale at default settings.)

Speed is very relative these days when it comes to ML based solutions because new hardware seems to scale that part way faster than regular compute. I think best way would be to provide three different pretrained models with different complexity so user can choose between speed vs quality.

Isn't Topaz like, 10x slower?

Is this maybe a "Be careful what you wish for" scenario?

(So far I've been impressed with the results of 2x SuperScale at default settings.)

Its okay, but doesn't match with competition