X265/HEVC Scaling divisible by 4/8/16? Macroblock/CTU question

[abaxas]

Hi all,

It's probably a newbie question but google has failed me.

In years gone by, it was normal (or often required) to scale any video so it was a multiple of 8 on both axis. Ie conform to a fixed number of macroblocks.

With x265/hevc we now have 16x16(+) CTUs so is it best practice to align the scaling to a multiple of 16? Or is 8 or 4 acceptable?

The reason for this is I'm doing ultra low bitrate encodes (<100kbps) of mostly static content and my mind thinks that using a multiple of 16 would provide for best efficiency.

Thanks

Abx

[foxyshadis]

It probably isn't worth it, because x265 will repeat the last line out to the internal border. It can't hurt to try, though, as the tiny overhead of coding even static blocks might add up at low enough bitrates, but you'd have to align to CU size, which is as high as 64x64.

[Asmodian]

In my opinion it is usually not worth the lost pixels, maybe at extremely low bitrates as foxyshadis mentioned. I think mod 4 is a good idea due to chroma subsampling but worrying about mod 8, 16, 32 or 64 isn't worth cropping extra pixels.

Even 4K isn't mod 64. :( ;)