10-bit vs 8-bit: Hype or Real?

[KnightsFan]

A decent test would be to shoot a scene in as high quality as you can, like uncompressed raw, and then export a 10 bit and 8 bit version with roughly matching codec and size from that Raw master, and compare those results. If you really want to isolate 10 vs 8 bit, export uncompressed videos with those bit depths. You will most likely see the biggest difference in scenes with smooth gradients in the shadow, particularly with big color grades such as incorrect white balancing, or underexposed scenes.

Maybe I'll do some tests later today.

[KnightsFan]

I did do a quick test. My process was to film a white wall as a 4k Raw clip, which I processed into a 4:2:2 10 bit uncompressed file. I then used ffmpeg to process the uncompressed video into two different clips with the only difference being the bit depth. I used 420 color and crf 16 on both. The two files both ended up roughly the same size. (8 bit is 3,515 KB, 10 bit is 3,127 KB).

I applied a fairly extreme amount of gain, and white balance adjustment equally to all clips. I've included a 100% crop of the uncompressed, 10 bit, and 8 bit files. As you can see, the 8 bit has significantly more ugly banding than the 10 bit.

As you can see, the 8 bit has some nasty banding that is not present in the 10 bit version. This is of course an extreme example to show a relatively small difference, but also it does get perceptually worse in motion rather than still frames. Also note that the PNG files themselves are 8 bit (which would match a typical delivery). The banding you see is from the color grading, as all 3 versions have been quantized down to 8 bit upon rendering.

Moreover, the 10 bit is actually a 10% smaller file. I find 10 bit HEVC is consistently a smaller file size than 8 bit for better quality. The real benefit of more accurate sampling is that it allows more accurate processing throughout, from compression to coloring.

On an related note, both the HEVC clips have lost all the grain and detail compared to uncompressed, which is very unfortunate. However, they are 1% of the file size so I can't complain too much!

 

 

Edit: just look at the file names to see which pic is which

[luckyluca]

On 1/21/2021 at 5:15 AM, kye said:

Here's the vectorscope that shows the 8-bit quantisation, which is made worse by the low-contrast lighting conditions and the low-contrast codec (C-Log)

I managed to "fix" the noise through various NR techniques, which also fixed the 8-bit quantisation:

Hi Kye,

thanks for your post, would you mind mentioning how you fixed these kind of 8bit banding/quantisation artifacts in post?

To give context, I'm shooting using an Iphone 11 which is 8 bit. I was wondering if 10bit alone justifies purchasing a brand new Iphone 12 pro max, compared to using 8bit. Pls note I normally shoot 8bit and expose to protect highlights, and do not too grade aggressively. 

Also, I just ordered the freewell vnd kit which includes a mist base. Would you expect to see more aggravated banding issues when using mist filters on 8bit footage?

 

 Best

L

[kye]

On 8/23/2021 at 1:15 AM, luckyluca said:

Hi Kye,

thanks for your post, would you mind mentioning how you fixed these kind of 8bit banding/quantisation artifacts in post?

To give context, I'm shooting using an Iphone 11 which is 8 bit. I was wondering if 10bit alone justifies purchasing a brand new Iphone 12 pro max, compared to using 8bit. Pls note I normally shoot 8bit and expose to protect highlights, and do not too grade aggressively. 

Also, I just ordered the freewell vnd kit which includes a mist base. Would you expect to see more aggravated banding issues when using mist filters on 8bit footage?

 

 Best

L

The treatments in the above (IIRC) were through applying noise reduction (NR).  It was a combination of spatial NR which essentially blurs the footage, and temporal NR which blurs between frames.  NR is quite smart, but ultimately it's a fancy blur, and definitely softens the footage visibly, so needs to be used with sharpening / unsharp mask, and has very significant limitations in real use.

In terms of 8-bit vs 10-bit and if it's "worth it", the test is really if you have problems with 8-bit footage.  If you're shooting 8-bit and not having banding issues, or issues with colour blotches on skin tones, etc, then moving to 10-bit probably won't have much impact.

My advice is to focus on the final results you're getting, work out what the biggest issues / limitations are for you, and then put your money towards those issues.  Counter to what these (and many other) forums suggest, most of the time the biggest return on investment for the average film-maker is on something other than their camera, most likely audio equipment, lenses, filters, audio equipment, lighting and grip, haze / fog, production design, and, to save the best for last, the best money you can ever spend is to invest in your own education and skills.  The skilled operator can get reasonable results in almost any situation with any equipment, whereas amateurs make Alexa footage look like a home video....

[John Matthews]

The way I look at this is:

1. If you have enough detail and no banding in 1080p, 8-bit 1080p is just fine (for me).
2. If you see banding (i.e. in the sky), shoot 4k 8 bit.
3. If something is still "off" (with colours), try 4k 10 bit and I doubt you'll see anymore problems.

I've never been at step 3 for my purposes, nor do I have any cameras that shoot 4k 10 bit.

[TomTheDP]

If shooting in LOG definitely go 10 bit. If shooting REC709 you won't really see a difference. My 2 cents 

It does largely depend on the camera though. CLOG is also a less flat log curve so it doesn't really benefit from as much data. 

[herein2020]

On 8/27/2021 at 4:05 PM, TomTheDP said:

If shooting in LOG definitely go 10 bit. If shooting REC709 you won't really see a difference. My 2 cents 

It does largely depend on the camera though. CLOG is also a less flat log curve so it doesn't really benefit from as much data. 

For me the answer is simple, I shoot 10 bit unless my camera only supports 8 bit. For example my drone only shoots 8bit for 6K but will do 10bit for all other resolutions. My S5 will only shoot 10bit with a 30min limit whereas it will shoot 8bit with no limit so if I need to run continuously over 30 min I would drop down to 8bit.

To me seeing the difference in output is nearly indiscernible except in certain situations. The real power of 10bit however is the ability to adjust the white balance to a greater degree without ruining the footage and the other real power is being able to lift the mids to a greater extent without losing color.

I shoot a lot of modelling and fashion work; usually very quick run and gun work with no supplemental lighting, being able to properly expose the skin while keeping the proper skin tones in high DR situations without supplemental lighting is where 10bit lets me lift the mids without losing skin tone color. With 8bit I have to leave the skin under exposed because there is no color in those mids if they start out under exposed.

For properly lit scenes or scenes where very little editing will be done; 10bit provides nearly no benefit over 8bit.

[Django]

FWIW, I was shooting 8-bit for years on Sony's and Canons and when finally upgraded to a FS7 and R6, the difference for both S-log3 and Clog/Clog3 was night & day. The latitude, the colours.. it all gets enhanced and doesn't break anymore. This has allowed me to up my grading skills in Davinci by being able not just to push things harder but go deeper using qualifiers and more advanced techniques.

To be honest, I don't even lust for RAW anymore and the associated storage headaches that go along with it. Chunky bitrate 10-bit can go a long way. And with the M1 Macs it's now a breeze to edit. I really hope Sony gives us a 10-bit option in A7 IV and other upcoming Alpha models. 8-bit is a thing of the past as far as I'm concerned and I've let go of all my 8-bit cameras..

[herein2020]

12 hours ago, Django said:

FWIW, I was shooting 8-bit for years on Sony's and Canons and when finally upgraded to a FS7 and R6, the difference for both S-log3 and Clog/Clog3 was night & day. The latitude, the colours.. it all gets enhanced and doesn't break anymore. This has allowed me to up my grading skills in Davinci by being able not just to push things harder but go deeper using qualifiers and more advanced techniques.

To be honest, I don't even lust for RAW anymore and the associated storage headaches that go along with it. Chunky bitrate 10-bit can go a long way. And with the M1 Macs it's now a breeze to edit. I really hope Sony gives us a 10-bit option in A7 IV and other upcoming Alpha models. 8-bit is a thing of the past as far as I'm concerned and I've let go of all my 8-bit cameras..

I feel the same way about the S5 and VLOG. The latitude that I have with the footage I thought was only possible with RAW or some form of compressed RAW. After you discover 10bit combined with a LOG profile and Davinci Resolve there is no going back. 

[hyalinejim]

I absolutely agree. 10bit 422 VLog is close enough to raw for me.

If you can set up your color correction that exposure, contrast and colour adjustments work consistently across the tonal range of the log curve (for example, Premiere's Lumetri doesn't, but an ACES workflow does) then it's effectively giving you the power of raw exposure, contrast and white balance adjustments without breaking the footage and maintaining a good degree of accuracy. But 8 bit would fall apart.