Thoughts on comparing 8-bit and 10-bit footage from the Panasonic GH5
Categories
Panasonic GH5

Thoughts on comparing 8-bit and 10-bit footage from the Panasonic GH5

Why isn’t the 10-bit footage in the Panasonic GH5 better than the 8-bit footage?

This video is specifically made to address a confusion many people seem to be having as to why there aren’t any significant differences between the 8-bit and 10-bit footage in the Panasonic GH5, as of March 2017.

I’ve tried to keep it as simple as possible:

You can use the same logic to check for differences in bit depths in other cameras as well.

The takeaway

  • For 8-bit 4:2:0, shoot 100 Mbps
  • For 10-bit 4:2:2, shoot 400 Mbps internally, or
  • For 10-bit 4:2:2, shoot 880 Mbps Prores HQ externally

Calculations and links from the video

The video references a few articles I’ve written earlier; and some calculations that will help you follow my arguments. Here are some preliminary articles that explain the terminology, formulas, etc., mentioned in the video:

  1. What is Video Compression?
  2. The Difference between Lossy and Lossless Compression
  3. What is color bit depth, color space, color model, CIE, luminance and chrominance, and chroma subsampling?
  4. What is inteframe and intraframe compression?
  5. What’s Stopping us from having a Universal Codec that will Benefit Humanity?
  6. Bits vs Bytes, and the story behind Disk Drive Sizes
  7. Bit Rate vs Data Rate

Calculations

I’m assuming you’ve read the articles above, because these calculations follow everything I’ve laid out earlier.

A full color image (produced by a camera) is a raster image and has three channels – Red, Green and Blue. That’s why we have to multiply the data by 3. If you’re shooting RAW, there’s only one channel, because RAW hasn’t been debayered yet. Once it is, it will also have three channels.

As a separate note, YUV footage also has three channels, one for luminance and two for chrominance. To learn more about this read the following articles:

  1. What’s the difference between YUV, YIQ, YPbPr and YCbCr?
  2. Don’t be Confused between RGB 444 and Y’CbCr 444

Step One: Uncompressed file sizes

As mentioned in the article above, to find the uncompressed file size of one frame, use this formula:

Channels x Bit Depth x Chroma Subsampling Percentage x Resolution

In the case of the GH5, here are the results for 8-bit and 10-bit footage in 4K (MB per frame):

8-bit size: (3 x 8 x 0.50 x 4096 x 2160) / (8 x 1024 x 1024) = 12.7 MB

10-bit size: (3 x 10 x 0.67 x 4096 x 2160) / (8 x 1024 x 1024) = 21.2 MB

We are dividing by (8 x 1024 x 1024) to convert bits to Megabytes. Also remember, there’s always a small overhead for file and metadata information, but it’s negligible for calculation purposes.

The next step is to get the data rate by multiplying by the frame rate:

Data rate = Size of one frame x number of frames per second

In this case let’s assume the frame rate is 24 fps:

8-bit size: 12.7 MB x 24 fps = 305 MB/s or 2438 Mbps

10-bit size: 21.2 MB x 24 fps = 509 MB/s or 4070 Mbps

So, you can see that 10-bit needs about 1.67 times the space as 8-bit to be different:

4070/2438 = 1.67

Step Two: What you get with the Panasonic GH5

The Panasonic GH5 gives you 8-bit 4:2:0 at 100 Mbps and 10-bit 4:2:2 at 150 Mbps internally. Later in 2017 we’ll have the 400 Mbps option.

How much compression is this? Here’s the math:

8-bit 4:2:0 – From 2438 Mbps to 100 Mbps is 24 times!

10-bit 4:2:2 – From 4070 to 150 Mbps is 27 times!

What you’ll see is that even though 10-bit needs 1.67 times more data, it is more heavily compressed, and therefore is only bigger by:

27/24 = 1.1 or 10%

To really maintain the difference, 10-bit will have to compressed by:

24.32/1.67 = 14.56 times

So, the data rate for 10-bit 4:2:2 should be at least:

4070/14.56 = 280 Mbps approximately

This is if it’s an interframe codec. For an intraframe codec, the number is double that.

Step Three: The broadcast specification

Read this first: Choosing a broadcast camera

The traditional way has been to equate interframe and intraframe in this way:

50 Mbps interframe = 100 Mbps intraframe (for 1080p)

4K is four times the data:

200 Mbps interframe = 400 Mbps intraframe (for 4K)

As we saw above, 10-bit 4:2:2 should have a data rate of 280 Mbps in interframe. But the 400 Mbps option will arrive as an intraframe codec. Therefore:

Necessary data rate: 280 Mbps x 2 = 560 Mbps

This is the data rate the GH5 needs in 10-bit 4:2:2 just to be better than 100 Mbps 8-bit 4:2:0. However, the GH5 will only have a 400 Mbps option internally. If you need more, you need to attach an external recorder and record Prores HQ, which has a maximum data rate of 880 Mbps.

Why can we only shoot 400 Mbps internally?

That’s something only Panasonic can definitely answer. Though one reason could be that, internally, you need a V60 class card to write 400 Mbps. V60 stands for 60 MB/s (Class 10 is 10 MB/s).

400 Mbps = 50 MB/s

As you can see, this is just under the maximum spec for V60. That’s probably why the GH5 comes in at 400 Mbps. To have a higher data rate, it needs a better class card (a V90 perhaps, which is 90 MB/s or 720 Mbps). Whether this is doable or not is Panasonic’s call.

As for 150 Mbps in 10-bit, forget it!

It’s not very wise to compare 8-bit and 10-bit in its present form. When the 400 Mbps version comes out, 10-bit will have the room to show if it really is better. Then, it will have no more excuses if it falls short.

Step Four: Color, another angle

You can also think of this in terms of color:

8-bit = 256 shades = 16.7 million colors

10-bit = 1024 shades = 1 billion+ colors

Difference = 1 billion / 16.7 million = 64 times

You’d think you need 64 times the data, but the reality is, the human eye can’t see more than 10 million colors – at best.

Read this: Colors of the Human eye.

Finding the exact number is impossible. On the other hand, the extra space in 10-bit is always useful in color grading. Read this to learn why: Advantages of working in 32-bit float

Overall though, I don’t think the number of necessary colors (both visual and theoretical) is more than what you get with 880 Mbps Prores HQ, so that’s a great upper limit to go by. If you want the most color juice, use an external recorder. How much of an improvement this is we’ll know once the firmware updates are released.

The takeaway, again

  • For 8-bit 4:2:0, shoot 100 Mbps
  • For 10-bit 4:2:2, shoot 400 Mbps internally, or
  • For 10-bit 4:2:2, shoot 880 Mbps Prores HQ externally

I hope this has helped in your understanding on how bit depth, chroma subsampling and compression works in cameras. What do you think?

Exclusive Bonus: Download your FREE Blueprint: How to make a movie. A complete visual representation of the filmmaking process from beginning to end.

26 replies on “Thoughts on comparing 8-bit and 10-bit footage from the Panasonic GH5”

I believe your conclusions are correct but your reasoning is incorrect.
I have analysed the clips that the GH5 produces at 100 Mbps 8 bit and 150 Mbps 10 bit and found that the codecs settings are different. The 100 Mbps uses full IBP while the 150 Mbps only has I and P frames and is less efficient. The net result is that the average I frame size is 1.57 MB for the 100 Mbps and 1.25 for the 150 Mbps so there is actually less information on the 422 10 Bit frames on average. It is possible that the GH5 does not have enough processing power to compress the 150 Mbps as it does the 100 Mbps, When you go into the All-Intra format the average frame size is 1.9 MB which is the best of the options and you have no motion interpolation errors. So looking purely at clips out of the camera it would seem All-intra > 100 Mbps > 150 Mbps. Having said that many people that use 422 can’t tell apart All intra from 150 Mbps…

I stumbled upon your site while researching a question I have on the impact of frame rate and data rates. Awesome article (and site).

If anyone reads this post, I’d like to figure out which recording format setting on my 4K camera will result in the best possible image quality. I understand the reason one may want to use different frame rates, but I’m not sure why my camera offers different bit rates which appear to be dependent upon the frame rates.

The formats are 4K/60P 150M, 4K/30P 100M, and 4K/24P 100M. All else being equal, would 4K/24P 100M capture the most data and therefore result in better image quality?

What are your thoughts on shooting v-log with the All-I 400Mbps 10-bit?
Does the new codec and bitrate fix the problem?

This is awesome – how does this compare with a sony fs7?

It’s 422 4k intraframe codec bitrate is 240mbps…
Doesn’t this 400mbps on gh5 far exceed, or am I missing something?

Hey Sareesh thank you very much for this! Reading this site is damn informative and i wish more people talked about these subjects as thruthfully as you do.
I have a question regarding Ext Recorders : do the 880Mbps (Atomos) ProRes already apply or do I have to wait for the firmware update ?

Hi Sareesh,
what is the conclusion about external recorders?
If the signal is compressed and then rescaled what is the benefit of external prores recorder?
Before I read this I thought that there were no compressions on the hdmi signal from the camera and that was just the point – to catch the uncompressed signal from the censor to the recorder.
Ken

It is uncompressed, but uncompressed like reopening a sandwich wrapped in aluminum foil paper.

Nice work. Maybe i am wrong but the calculation seems incorrect. I think your calculations is a factor 1.67 too high.

The 10 bits datarate is 4070 Mbps.
10-bit size: 21.2 MB x 24 fps = 509 MB/s or 4070 Mbps

Suppose we are happy with the same compression as 8 bits at 100 Mbps, that is 24x compression (8 bits = 2438 Mbps to 100 Mbps is 24 times)

The 10 bits datarate is 4070 Mbps / 24x compression = 170 Mbps.

So 170 Mbps is enough for a 24x times compression which should be enough.
Please correct me if i am wrong.

You really get things well explained and even for guys that work in the area, it is awesome to view your content!

Really awesome stuff Wolfcrow. Keep it coming

Thanks for the review.
In 1920×1080, 24p, 200Mbps: 10 bits, 4:2:2 All-i will be good internal, am I wrong?
An we’ll use an external rec for 4K.

Sareesh I was wondering how you view has changed regarding the GH5s That is meant to be more tailor made for video. Id like to use it on a broadcast documentary for the BBC Ill be shooting in Kashmir next month. .. The rendering time is not an issue but the maount of material id be eating up cards at 400mbps. Does the gh5s record ten bit better now?

Are you really sure that the HDMI output on the GH4/GH5 is just an output of the compressed signal? Are there been any test to verify this?

The signal is uncompressed, but for the GH4 and Sony a7 series the output has already been processed and then uncompressed to meet HDMI specs. This is what I’ve seen through my tests.

Hi, this is a very detailed video, but you re unfortunately completely leaving behind the real advantage of 10 bit right now available, which is the possibility to finally use vlog efficiently in camera and being able to grade the footage without banding artefacts. in your resume it sounds as if the advantage of 10 bit color would be useless right now but its absolutely nececary and useful from start for working with log footage. For more information check out these videos and connected articles: https://www.slashcam.de/artikel/Test/Panasonic-GH5—8-Bit–10-Bit–4-2-2–V-Log-L—alles-OK—-Die-GH5-in-der-10-Bit-Praxis.html
https://www.youtube.com/watch?v=RBjQojrvCg8
https://www.youtube.com/watch?v=agA9796bs3U

Nobody said it isn’t beneficial. The point is the benefit isn’t there with heavy compression.

i see great benefit in all the examples. the benefit is beeing able to use vlog in camera, which benefits in an extra stop of dynamic range. and it looks good to me, and very usable. this whole heavy compression talk is getting towards pixel peeping. Pro res lt is also too heavy compressed regarding this math, but i doubt anybody here can really tell lt and the normal 422 or hq apart in real life examples.

What would you transcode the gh5 files to to
Edit in fcp x if shooting 10 bit 422 internally?.. prores or prores hq?

Comments are closed.