x265 10 bit encoding from 8 bit source

[Gurn]

I keep seeing claims that using 10 bit x265 encoding from 8 bit sources is optimal.

Is there any source that offer some solid justification for this claim?

I personally am seeing the opposite, slightly larger encodes and micro stuttering consistently for moments after a video starts playing.

Playback testing on VLC and various Roku players.

This is based on using the Nvenc codec in Handbrake 1.10.2 with these settings:

Audio passthrough: MKV 1080 File contains a E-AC-3 audio track

FPS: Same as source
CFR
Color Range: Same as Source
Encoder Profile and Level: Auto
CQ 28

Tia

[Z2697]

nvenc is no x265
stutter has nothing to do with bitdepth, unless your gfx card do main but not main10 and you cpu is very weak so decoding is slower than playback
but if your nvenc can do main10 shouln't your nvdec can do it as well?

[GeoffreyA]

In theory, it's supposed to help with banding.

[Z2697]

It helps when the source have dithering and without it the source would have significant banding.
(non-post-8bit-quantization grain/noise can have the effect of dithering)

If the dithering pattern is destroyed by lossy compression, the "fall down" value can be inbetween of one 8-bit code level...

So, if your source relies little on dithering, or your bitrate is high enough, 10 bits encoding will help but not much.

[RanmaCanada]

You're using NVENC, which is hardware encoder and is NOT x265. You also failed to mention what generation card you are using, which is extremely relevant as earlier renditions of NVENC didn't even support b-frames. Your main problem is it is NOT x265 as mentioned. 10 bit encoding helps remove banding, especially so on anime. It also helps with blacks.

Hardware encoding takes multiple generations to reach software encoding quality. It was only recently that hardware encoding to h264 could reach the fidelity of x264, a 20 year old encoder.

Rigaya has a great chart (https://rigaya.github.io/vq_results/) to show you just how far behind nvenc is in comparison to x265. Blackwell still can't reach x265 very slow and Pascal is a joke.

[GeoffreyA]

And on dark gradients in anime, aq-mode 3 may be needed as well.

x264 had a significant efficiency gain when using 10 bit; it is less pronounced with x265.

[microchip8]

And on dark gradients in anime, aq-mode 3 may be needed as well.

x264 had a significant efficiency gain when using 10 bit; it is less pronounced with x265.

That's 12 bit for x265...

[Z2697]

I wouldn't think 12bit x265 has much of a gain, or my eyes are not good enough for the amount of gradients.

[GeoffreyA]

That's 12 bit for x265...

Somehow, reminds me of the quote, in relation to Penrose's theory: "the shortest path between two truths in the real domain passes through the complex domain."

[microchip8]

I wouldn't think 12bit x265 has much of a gain, or my eyes are not good enough for the amount of gradients.

I did a few encodes in 12 bit with a problematic source for 10 bit and there was quite an improvement. But that was many ages ago, since then I haven't touched 12 bit. Also, compatibility, etc

[DJ Bobo]

I think it is bad tuning of the encoder. 10-bit encoding should NOT yield better quality than 8-bit at the same bitrate, but somehow it does.
I tried it, it's true, there is significantly less banding with 10-bit, even from 8-bit sources. And using the same CRF parameter, the 10-bit result is even smaller than the 8-bit one, which is totally counter-intuitive, and not logical at all.
So again, the problem is bad behaviour of x265 in 8-bit mode. It drives me nuts to see perfect gradiants on the original DVD, and then get severe banding in 8-bit mode at 1500kbps, even though CRF18 yields 700kbps!!! Crazy!

[excellentswordfight]

I think it is bad tuning of the encoder. 10-bit encoding should NOT yield better quality than 8-bit at the same bitrate, but somehow it does.
I tried it, it's true, there is significantly less banding with 10-bit, even from 8-bit sources. And using the same CRF parameter, the 10-bit result is even smaller than the 8-bit one, which is totally counter-intuitive, and not logical at all.
So again, the problem is bad behaviour of x265 in 8-bit mode. It drives me nuts to see perfect gradiants on the original DVD, and then get severe banding in 8-bit mode at 1500kbps, even though CRF18 yields 700kbps!!! Crazy!
This is my experience as well, when I did early tests, it wasnt just banding, there was overall improvements all over the place with main10.

This is a frame from a an encode from a 8bit source of ToS, only change in the encode is 10bit vs 8bit.

8bit
https://ibb.co/7Qbxxx5
10bit
https://ibb.co/ChLFrTB

So most abvius is less banding on flat areas/gradiants. But also look at the green section at bottom, 10bit is keeping more details there.

[Z2697]

It helps when the source have dithering and without it the source would have significant banding.
(non-post-8bit-quantization grain/noise can have the effect of dithering)

If the dithering pattern is destroyed by lossy compression, the "fall down" value can be inbetween of one 8-bit code level...

So, if your source relies little on dithering, or your bitrate is high enough, 10 bits encoding will help but not much.

this.