10bit x264 CRF vs 8bit x264 CRF

[meshaun]

I came to know that 10bit x264 saves more quality in a smaller file. And, if so, is there a slight comparison between the CRF values, if anyone knows it?
Like :

- 18 10bit kinda equal to which in 8bit CRF?
- 22 10bit is equal to which in 8bit CRF?

Plus, when encoding a 720p using a Full BD, which is the better CRF value to use in 10bit x264?
I mean, I also want to preserve some good quality and also have a decent small filesize. I tried 22 CRF in the new Katy Perry-Part of Me but resulted a 2.1 GB in 10bit even.

So, is there any recommendation and also help?

[Selur]

if I remember correctly:
8bit quantizer range is from 0 to 69
10bit quantizer range is from 0 to 51

Assuming the crf range is the same as the quantizer range:
- 18 10bit kinda equal to which in 8bit CRF?
18 *51/69 ~ 14
22 10bit is equal to which in 8bit CRF?
22 *51/69 ~ 16

guess Dark Shikari and others that are more knowledgeable regarding the inner workings&co of crf can probably say more,..

[Dark Shikari]

The CRF scale is the same for both 8-bit and 10-bit (it goes lower for 10-bit, but the scale is the same). It is not the same as the quantizer range.

[meshaun]

So, I mean, 10bit holds more quality right?
If so, I can use a greater CRF like 24 in 10bit x264 and achieve quality in about 22 CRF in 8 bit? It was just an assumption, but is it like that?

[mandarinka]

Most likely not, 10-bit encoding isn't that powerful.

[the_weirdo]

So, I mean, 10bit holds more quality right?
If so, I can use a greater CRF like 24 in 10bit x264 and achieve quality in about 22 CRF in 8 bit? It was just an assumption, but is it like that?

It also depends on source. For example, if source is 2D anime with a lot of gradients and little or no grains, then CRF 24 in 10-bit may have similar quality with CRF 22 in 8-bit, but on live-action materials then it's likely not.

[jpsdr]

For watching a lot of anime fansub releases, i confirm the_weirdo. For the special case of 2D anime without noise, on same file size, 10bit produce better result than 8bit. After, only dev (like DS) may confirm the fact that it's probably only a special case.

[Keiyakusha]

less banding may be a special case but this is pretty much main case why 10bit is interesting. depending on the source and the viewer this may have very different level of importance. (I hardly see any benefits of 10bit in live actions, unless the sky is shown there pretty often.) From my experience by reducing crf by 2 points (edit: or actually increasing, since more=less quality), 10bit may still save you from banding, but quality of the more visible parts of the image (that is seen by anyone without much effort) will drop just as much as in 8bit.
I don't think it is very good idea to compare 8bit to 10bit at all. I think of them as different ways of encoding stuff, like different formats (well, in fact they are different formats). For different types of sources I have different options for both 8 and 10bit encoding, but none of them based on each other, they developed from the ground by testing and looking if result turned out good enough. And for me it turned out that crf values are not that differ, but other ratecontrol options and the way I process source does.

[the_weirdo]

[...] And for me it turned out that crf values are not that differ, but other ratecontrol options and the way I process source does.

I mostly argee with you, while 10-bit does help a lot in reducing banding, quality of other parts is still degraded when using higher CRF value than with 8-bit. (I wish most fansubbers in my country also understood that). Maybe because I used to encode anime (720p) at crf 16 with 8-bit x264, that why I don't see much difference when compare with 10-bit at crf 18. Also, like you said, depending on sources, we should use appropriate options for 8-bit and 10-bit encoding instead of basing on each other.

However, when you're comparing two encodes with different input sources (the uncompressed raws that feed directly to x264 encoding engine, not the unprocessed sources) and different settings then that comparison may not be valid to judge the effect of different crf values.