View Single Post
Old 11th January 2019, 10:42   #6618  |  Link
excellentswordfight
Lost my old account :(
 
Join Date: Jul 2017
Posts: 322
@Ma

I asked some questions regarding mereange a while ago but didnt get a reply, so I thought that I give it another shot.

I've seen discussions here and in other threads that CTU 64 is overkill for resolutions bellow 4k, and I've seen from my own testing that lowering CTU from 64 to 32 on systems with plenty of threads gives an speed improvment of up to 50% for 1080p video, and lowering merange gives another 10%. And this is with a very minor compression hit. This behavior is also stated in this document https://media.readthedocs.org/pdf/x265/default/x265.pdf when it comes to threading performance.

What I find a bit odd is that this is stated in the document: "Given these considerations, you can understand why the faster presets lower the max CTU size to 32x32 (making
twice as many CTU rows available for WPP and for finer grained frame parallelism) and reduce --merange"
and this: "The default is derived from the default CTU size (64) minus the luma interpolation half-length (4) minus maximum subpel distance (2) minus one extra pixel just in case the hex search method is used."

But I cant see that any preset changes the merange value of 57, even the two fastes ones that do lower the CTU value to 32. How come? And since lowering CTU (and merange), can have such massive influence on speed, wouldn't be a good idea to have these values set based on resolution?

And giving the explanation of the default merange value, would the same calculation stand when lowering CTU to say 32? I.e. would that give an "best practice" value of 26 if me star is used?

Last edited by excellentswordfight; 11th January 2019 at 10:50.
excellentswordfight is offline   Reply With Quote