Differences between Sony 50M 4:2:2 10bit and 100M 4:2:2 10bi

Apologies for a slightly off-topic post, but I hope someone may have an answer. I just purchased a Sony A7Cii; I'll be recording in XAVC HS 4K (Sony's flavor of H.265), and am trying to decide between recording in 50M 4:2:2 10 bit and 100M 4:2:2 10 bit. I'd prefer to get the best image quality possible, but I'm starting a long-form documentary that will probably require a lot of footage and my storage is not unlimited.

What sort of testing would best reveal the quality differences between the 50M and 100m recordings? Should I be looking for resolution and noise differences? Motion artifacts? Degradation when pushing color grades to the limit? I guess another way of asking this is should I shoot some lens charts at different ISOs, or shoot some sort of test that would involve (repeatable) motion, or play with some crazy color grades—or all of the above? Any suggestions appreciated!

Resolve 18.1.4 Studio (soon to be upgraded to 19 or 20), M1 Mac Mini 16GB ram, Mac OS 12.7.6

My goto test for lossy codecs has always been high-motion high-details scenes. Leaves blowing in the wind, water splashing around, that kind of thing. I'm looking for encoding artifacts.

They're both 4:2:2 10 bit, so I would expect no difference in color.

One thing you didn't mention was the GOP structure for each. That could matter.

For 4K? How much shooting are you going to do that storage is an issue?

But less things like leaves moving in the breeze. If OTOH you're doing static shots with the camera locked down you should test locked down and likely not see any issues.

Sony‘s encoders are really good, in most situations you won’t see a difference.
What Jim describes are good test cases, I’d like to add the surface of water in backlight on a windy day. All covering full screen, just like the moving leaves.
Don’t look at the image only, the waveform can be revealing too.

Let's get some rough data involved:
50 Mbit/s = 6,25 MB/s.
100 Mbit = 12,5 MB/s.

A 1 TB ssd (my personal favorite is DC600M) can store the following
960GB*1000(for MB)/6,25(MB/s)/60(for minutes) = 2560 minutes or 42 hours at 50 Mb/s
960GB*1000(for MB)/12,5(MB/s)/60(for minutes) = 1280 minutes or 21 hours at 1000 Mb/s

A 4TB would quadruple the total stored length to 168 and 84 hours respectively.

Remember backups! I like backblaze but any place that fulfills the 3-2-1 (3 copies of your data, 2 different types of storage, 1 offsite) backup rule is worth it.
--------
It is my personal experience that I have been extremely sad when I didn't use the maximum bitrate to capture a thing.

Well, I'd recommend going as high as you can too, since 50 is not too much even for HD (and many cameras are delivering that). The data rate should be quadrupled for UHD for the same quality. Sony is not offering that, probably because of the high prices for faster media. These are all marketing considerations.

BTW, I can get close to 200 for HEVC in UHD on my iPhone. You can see waveform comparisons in this article:
https://digitalproduction.com/2024/06/1 ... m-the-fon/

CFe B cards aren't that expensive today. I've bought gen 4 Prograde 1TB cards for I think less than €400. I think only one camera in the world even supports Gen 4.

Sony uses those small cards so they can make the cameras smaller.

If you're doing static shots with the camera locked down you should test locked down

Not sure that would stress the encoder enough to be a useful test. It'd be like testing a race car's new high performance engine by driving down the street at 30 mph.

You need to get that thing on a race track, open her up.

My feeling is you should practice the way you play.

I agree your suggestion will show any problems far better but if the OP is doing low movement stuff does it matter?

My feeling is you should practice the way you play.

For getting better at what you do? Sure.

For testing new equipment? I don't think that's useful. You need to push that equipment to the breaking point, otherwise you could inadvertently go past that breaking point on set and end up with a problem.

It's always good to know the extent of your equipment's capabilities, even if you don't plan on exceeding them.

Jim is correct about testing certain scene types. That's why researchers use a collection of scene types representing a wide variety of encoder challenges:

Xiph encoding test material: https://media.xiph.org/video/derf/
UVG Dataset: https://ultravideo.fi/dataset.html
Whitepaper on UVG Dataset: https://web.archive.org/web/20240619113 ... _ready.pdf

You don't want to just eyeball differences. It's best to use objective image evaluation tools, then confirm it with your eyes.

Most of these tools require a reference video that must match frame-by-frame to the higher compression file. For a camera file test, that really requires simultaneous recording using two different codecs. I've done that many times using an Atomos Ninja V or Shogun 7 recording ProRes 422, then comparing it to the simultaneously recorded in-camera codec.

However, if the question is image quality of a Resolve-encoded file vs the original file, that only requires a single original file since you are encoding the higher compression file from the original.

On MacOS, Apple provides a free command-line tool called AVQT. It is designed to measure perceptual quality loss, not just a technical parameter. It can be downloaded from the Apple Developer site but that requires you create a free developer account (click "Account" at top-right corner of this page: https://developer.apple.com/.)
WWDC21 talk on AVQT: https://developer.apple.com/wwdc21/10145
Youtube tutorial on AVQT:

You can also measure visual generation loss using a tool developed by NetFlix called VMAF (Video Multi-Method Assessment Fusion). Unlike AVQT, it is available on both Mac and Windows. VMAF is built into certain versions of ffmpeg. On Mac, ffmpeg-VMAF is available through the Homebrew package manager: https://brew.sh

On Windows, pre-built ffmpeg-VMAF binaries are available here: https://www.gyan.dev/ffmpeg/builds/

There is a VMAF web tool, but it only works with 1080p and has other limits. The command-line version is much more capable:
https://vmaf.dev

Thanks all for the very helpful replies! I agree that for testing new equipment max stress tests are useful - one never knows what the gear may need to be used for a year from now. Per Jim's and Uli's suggestions, I shot some footage of wind-blown foliage at both 50M and 100M. Quickly viewing the footage at 100% in a 4K timeline in Resolve both in motion and frame by frame I couldn't seen any artifacts in either file, or any other differences between them. I'll do a more thorough viewing when I have time this weekend.

I'll also take a look at the tools suggested by Joe as soon as I have time.