X265 Veryslow no better than slow???

[YaBoyShredderson]

I tested out several different crf values with the veryslow preset, and found that crf 20 gave me an average bitrate of about 7 accross different kinds of content. I then found the clips that had a bitrate around the average and encoded them with 2 pass 7mbits, with slow, slower and veryslow. I used the staxrip video comparison tool, and as far as i can twll, there is essentially no difference between them? Did i do something wrong here? There is a difference in the background noise, but its mot better or worse really, just a bit different flipping between the frames. Only in a few frames did i actually see a very very minor difference in small details, a single hair being a bit less blurry etc. Have i done something wrong or is veryslow really not that much more efficent than slow? Is it maybe because of the high bitrate?

[benwaggoner]

With x265, using higher presets doesn't reduce the bitrate nearly as often as it increases quality. At the same bitrate, slower presets will look better. But sometimes a slower preset with CRF can increase bitrate AND quality, as new tools find new places to spend bits to improve quality.

[YaBoyShredderson]

Originally Posted by benwaggoner

With x265, using higher presets doesn't reduce the bitrate nearly as often as it increases quality. At the same bitrate, slower presets will look better. But sometimes a slower preset with CRF can increase bitrate AND quality, as new tools find new places to spend bits to improve quality.

I used abr to test different presets at the same bitrate. In a normal encode i would use crf 20, averaging around 7mbits.

[benwaggoner]

Also, for typical continuous tone film and video content, veryslow and slower aren't going to look that different. It'd take a pretty big double-blind comparison to get statistics on the quantitative difference; there won't be much of a qualitative one since the same tools are generally used, just less aggressively in slower. Qualitative difference are more visible between the faster presets. Slow to slower can look different in a way that slower to placebo rarely does.

However, for atypical content qualitative differences can be seen. For example, in lossless compression, placebo can have a bigger jump in reduced file size from veryslow than from slower to veryslow, which is the opposite for normal encoding. And things like graphics and text can benefit more from more exhaustive use of tools.