Quote:
Originally Posted by Asmodian
I am getting a huge performance hit when using NNEDI3 quadrupling and any image enhancement set to "refine the image after every ~2x upscaling step".
e.g.
NNEDI3 quadrupling and AdaptiveSharpen with "refine the image only once after upscaling is complete", 28.5 ms
NNEDI3 quadrupling and AdaptiveSharpen with "refine the image after every ~2x upscaling step", 54.4 ms
super-xbr quadrupling and AdaptiveSharpen with "refine the image only once after upscaling is complete", 16.3 ms
super-xbr quadrupling and AdaptiveSharpen with "refine the image after every ~2x upscaling step", 17.2 ms
GPU clock speed was constant throughout this testing.
Is this large of a difference expected? I understand there is extra work when moving in and out of YCbCr for NNEDI3 but this is larger than expected. The hit from the NNEDI3 64 quadrupling itself is less than that from running AdaptiveSharpen with "refine the image after every ~2x upscaling step".  |
I haven't really done benchmark tests for these things, but obviously running those algos multiples times costs performance. My future plan is to try to modify/tune the sharpening algos to only have to run once after upscaling, but still achieve a good effect, regardless of the upscaling factor. Currently if the upscaling factor is very large, most of the sharpeners (except SuperRes) lose their effectiveness.