8bit vs. 10bit

[Ed Przyzycki]

This thread might need to be shifted around a bit to a different forum, but I'll start it in the XDCAM-EX forum, because that's where the footage originates. I'm just trying to figure out what editing workflow (specifically preview monitor type, intermediate codec, etc) will work best for my needs.

1. Is the HD-SDI out of the EX line 4:2:2, 8 bit or 10 bit? (assuming I am recording live out)
2. And, the compression to the SxS cards are 4:2:0, 8 bit, correct?
3. NanoFlash is still 8bit, correct?
4. A standard HDMI cable can handle 8, 10 or even 12bit, right?
5. But standard BluRay spec is only 8bit, right? (4:2:0 YUV) Or are they doing some as 10bit now?
6. Are some consumer screens now 10bit?
7. Are any broadcasters actually requiring 10bit anything? (other than camera source?)
8. So, how many of you are editing on a 10bit timeline with this content (i.e. Cineform), and what's the benefit? (I get the whole color grading and keying graphical elements, especially the possibility of seeing color banding when doing extensive correction in 8 bit - but does 8 bit degrade THAT much?) - Even things like FirstLight are 8bit at the moment...
9. Unless I'm editing for 2k, 4k (film, etc - which I am NOT) do I even need a 10bit editing environment if most of my source is in XDCAM-EX, and it either outputs to BluRay or the web?

It seems to me that the greater file sizes and re-renders when outputting to H264/Bluray doesn't seem to be worth it, considering the original footage type.

I did read these threads for reference. (http://www.dvinfo.net/forum/converge...-thoughts.html and http://www.dvinfo.net/forum/cineform...supported.html)

I appreciate your thoughts, and feel free to link to other articles I might have missed as I continue to wrap my head around this.

[David Heath]

A few weeks ago my reaction would have been "of course 10 bit must be far better than 8 bit!" Now, I'd have to add a lot of caveats. I posted recently on exactly that topic in this thread - http://www.dvinfo.net/forum/panasoni...ntra-p2-2.html (post no 19). To quote from that:

One point that was made to me recently is that 10 bit *recording* doesn't have much point IF THE NOISE LEVEL OF THE SOURCE IS MORE THAN THE BITDEPTH OF THE CODEC. Apologies for capitals, but it's a highly significant fact. And one which, I confess, hadn't occurred to me until I was very recently corrected.

In practical terms, what it means is that 10 bit recording is very valid for the high end cameras - the two least significant bits may be faithfully recording real information. Move down the scale, and any camera with less than 2/3" chips (and quite a few with 2/3"!) will have a worse signal-noise ratio. High enough noise to swamp the differences between an 8 and a 10 bit codec. So AVC-Intra 100 may be very worthwhile due to it's 10 bit nature in a camera like the 3700, but not show any practical difference between it and an equivalent 8 bit in a far more noisy camera like the HPX300. In the latter case, all 10 bit is doing is wasting 20% of the 100Mbs ........

[Barry J. Anwender]

Ed, I'll try to fill in some of the blanks:

1. Is the HD-SDI out of the EX line 4:2:2, 8 bit or 10 bit? (assuming I am recording live out)

This was a hotly debated topic when the EX-1 arrived and Sony finally made a formal announcement at the NAB when the EX-3 came on the scene. The entire EX line delivers 10-bit 4:2:2 from the HD-SDI output.

2. And, the compression to the SxS cards are 4:2:0, 8 bit, correct?

Correct, the SxS ports on the entire EX line record 8-bit 4:2:0 long-gop.

3. NanoFlash is still 8bit, correct?

Correct, Convergent is using the same 8-bit MPEG encoder chip (actually two) that resides in the EX cameras. However, I believe they are delivering 4:2:2 rather than 4:2:0.

4. A standard HDMI cable can handle 8, 10 or even 12bit, right?

HDMI 1.3 is specified to deliver up to 36-bits of colour, therefore capable of handling 12-bit sources.
HDMI 1.4 is specs to deliver up to 48-bits of colour infomration.

5. But standard BluRay spec is only 8bit, right? (4:2:0 YUV) Or are they doing some as 10bit now?

The MPEG portion of the Blu-ray spec is 8-bit for backward compatibility with legacy MPEG-2 delivery.
The early Profile 1.0 Blu-ray players like the early HD-LCD displays used 8-bit processing systems. Several generations have followed 1.1, 2.0 and now 3.0. As with display technology, your going to find a mixed bag with anything between 8-bit and 12-bit processing depending upon the manufacturer and the price of their product lines.

6. Are some consumer screens now 10bit?

My Pioneer Kuro 50" Plasma display, now two years old, complies with HDMI 1.3 and supports 12-bits throughout its entire processing chain to the physical display. Sony a year ago was using 10-bit processing for most of its LCD displays. I don't know where they are this year.

7. Are any broadcasters actually requiring 10bit anything? (other than camera source?)

Broadcasters in HD studios would be working with a minimum of 12-bits and more likely 14-bits in their HD-SDI chains.

8. So, how many of you are editing on a 10bit timeline with this content (i.e. Cineform), and what's the benefit? (I get the whole color grading and keying graphical elements, especially the possibility of seeing color banding when doing extensive correction in 8 bit - but does 8 bit degrade THAT much?) - Even things like FirstLight are 8bit at the moment...

The consideration on this question depends on the number of editing generations and/or type of editing or grading you plan to do. I have used 10-bit RAW RGB for some smaller projects where considerable editing is needed. The downside with RAW 10-bit is that you need a very fast RAID array and a lot of CPU time. I edit with FCP Studio 3.0 and for most projects my preferred workflow is 10-bit ProRes.

9. Unless I'm editing for 2k, 4k (film, etc - which I am NOT) do I even need a 10bit editing environment if most of my source is in XDCAM-EX, and it either outputs to BluRay or the web?

If your delivering to the web, then 8-bit long-gop is more than enough quality. Blu-ray H.264 on the other hand, will clearly benefit from 10-bit 4:2:2 source footage that comes out of the EX HD-SDI port. To this end, Black Magic Design is due to ship their H.264 Pro-Recorder which accepts HD-SDI 4:2:2 and uses 10-bit real-time encoding to H.264 output on a USB 2.0 cable. Perfect for the EX HD-SDI port. I am expecting delivery of this device any day now from B&H and look forward to putting it through it's paces for Green Screen work and Blu-ray delivery.

This is my experience to date and I am sure that others can provide more precision that I am "not" able to at this time. Lot's to learn but do learn by doing - Cheers!

[Steve Kalle]

6. Consumer displays 10bit?

NONE are 10 bit. Some might have 10 or 12 bit processing but that has nothing to do with the actual display and the number of colors the screen can display. For example, Dell's Ultrasharp 24" 2010 model has 12bit processing but only an 8 bit screen. My Eizo CG243W has 16 bit processing and a 10 bit screen.

Unless you are doing some serious grading and/or effects, then 10 bit will provide no benefit.

Have you looked into the Cinedeck? The first version 'Extreme' has been released with lower end models coming soon which will give 10 bit Cineform 422 recording.

About broadcast: several broadcast TV shows have been using the nanoFlash and/or XDR - both are 8 bit.

I was just next to all the Chicago news cameras and only one was 10 bit (2 were HPX170s, 1 - Betacam SX, 1 - 10 bit Panny, 1 - unknown)

One last point: the BBC standards do not require 10 bit at all.

[Ed Przyzycki]

Thanks - Barry, David, and Steve. I'm realizing very quickly that now more than ever I have to more clearly define my business needs and look very closely at what my clients are asking for now and in the future. (If they even know) When you are buying workstations and considering workflows, you want to plan for the future (i.e. 3D- ugg) and also provide some solutions that your competitors' can not (and also build your overhead with some longevity in mind). But on the other hand, I can't afford to invest in things just because they are 'the greatest'.

Thanks for the news on the H.264 Pro-Recorder. (Now I wonder if I need a dedicated internal capture card at all now)

Also thanks on Discovery and BBC info.

I figure with the EX3 (either to SxS or SDI) and a 10-bit intermediate (Cineform), that's good enough (for now).

[Richard Crowley]

"In the latter case, all 10 bit is doing is wasting 20% of the 100Mbs"
It is actually worse than that. Noise takes up useless bandwidth when compressing, further reducing the quality of even the non-noisy parts of the image (if any).

[Geoff Addis]

Surely the number of bits defines the theoretical dynamic range rather than the noise floor whilst the size of the image sensors and input amplifiers define the noise floor or am I wrong?

Geoff

[Tim Kolb]

Originally Posted by Geoff Addis

Surely the number of bits defines the theoretical dynamic range rather than the noise floor whilst the size of the image sensors and input amplifiers define the noise floor or am I wrong?

Geoff

The noise floor of a given camera is independent of any of this 8 bit vs 10 bit discussion...as is dynamic range.

8 bit vs 10 bit is strictly the color precision in the file. does it have 256 steps in each channel...or 1024?

Obviously the more precision we get, the more 'accurate' we are in showing the source material.

However, where 0 is and where 256 is in the sense of dynamic range is a virtue of the camera's imaging system...

(...and frankly our industry's 'shiny object' thirst for resolution is grossly misplaced...a sensor with bigger photosites or "cels" or "pixels" or whatever you want to call them, will almost always have better dynamic range than a sensor chip with more pixels which have to be smaller to fit on the same chip. Remember Canon's image aesthetic premium in the DV days? They had the lowest sensor resolution of nearly any DV camcorder available at the price range, but the dynamic range was superior, and the images looked great. It's why an F23 with 2/3" HD video camera sensors has greater dynamic range than a RED One with a massive 35mm image target...and 4K res)

8 bit color precision does more 'rounding' to create its image than 10 bit does...single pixel odd deviations can be lost in the rounding...10 bit can sometimes 'rescue' them.

I know of almost no image post production pipeline in broadcast that is running at 12 bit or higher color precision unless you're talking about a DSP processing step in a camera, or some rather arbitrary estimate of what would be an analog pipeline.

16 bit DPX and Open EXR workflows are out there in feature film post production, but the container is 16 bit because you have 8, 16, or 32 bit float available as a file structure...often 10 bit images will reside inside, what has to be for proper processing, a 16 bit 'container'.

[Steve Kalle]

I can testify to the benefit of editing with a 10 bit LCD and Photoshop. Just last week, I was asked to fix and edit an image that had artifacts from poor up-sizing. Moving the Photoshop interface between my Eizo and my Dell, there were color shades that I could not see on the Dell that I could easily see with the Eizo. Furthermore, the desktop image from HP that I use has banding on the Dell but not the Eizo. FYI, I use a Quadro FX3800 and its Displayport (10 bit) to the Eizo, and Premiere CS5 can output YUV to RGB realtime; thus, I don't need a pro I/O card like the Kona just to monitor.

So, while I don't think its absolutely necessary to capture in 10 bit, I do see a benefit to editing in 10 bit.

If you haven't done so, getting a high quality LCD for accurate color greatly helps especially because you output to BR. The Eizo is only $2k at B&H and the X-Rite calibrator is only $200. The cheapest 10 bit video card is the ATI 4870 with Displayport but if you edit with Adobe or Avid, then the cheapest is Quadro FX1800 ($450). BUT...David from Cineform mentioned getting 10bit via DVI with nvidia for Cineform ONLY. Technically, his response to my question about 10 bit & DVI was, "Under NDA".

[David Heath]

Originally Posted by Tim Kolb

The noise floor of a given camera is independent of any of this 8 bit vs 10 bit discussion...as is dynamic range.

8 bit vs 10 bit is strictly the color precision in the file. does it have 256 steps in each channel...or 1024?

Obviously the more precision we get, the more 'accurate' we are in showing the source material.

I disagree Tim. The noise floor of a camera front end is highly significant in any "is 10 bit worth it?" discussion.

Let's imagine a uniform part of the image - a nearly perfect mid-grey. A perfect noise free sensor would feed it to an 8 bit codec to yield uniform values of 128 - half way between 0 and 255. Feed the same to a 10 bit codec, and the improved resolution may tell you that it's not 4x128 - but 4x128 +2 - the 10 bit codec has defined the level of the mid-grey with more precision.

But now what if the sensor is not perfect, but has a given level of noise? Feed that to the 8 bit codec, and you may get a series of values such as 127, 128, 126, 129 etc. Feed it to the ten bit codec and you may get 127x4,128x4(+1), 126x4(-2), 129x4(-1) etc - in this case, most of what the ten bit codec is doing is more precisely defining the noise! The noise is effectively masking the slight genuine (and potentially useful!) variations that a 10 bit codec could resolve, and effectively making the use of a 10 bit recording codec academic.

Hence what I originally posted - "...........10 bit recording is very valid for the high end cameras - the two least significant bits may be faithfully recording real information. Move down the scale, and any camera with less than 2/3" chips.........will have a worse signal-noise ratio. High enough noise to swamp the differences between an 8 and a 10 bit codec."

Using 10 bit for post is a different matter, even if acquired in 8 bit.

[Tim Kolb]

David,

You and I are in lock-step agreement...and i mean completely.

I was trying to delineate these factors from each other...

Geoff asked:

"Surely the number of bits defines the theoretical dynamic range rather than the noise floor whilst the size of the image sensors and input amplifiers define the noise floor or am I wrong?"

My answer was an attempt to say that these factors are all determined separately...noise floor, dynamic range, color precision.

When I said:
"8 bit color precision does more 'rounding' to create its image than 10 bit does...single pixel odd deviations can be lost in the rounding...10 bit can sometimes 'rescue' them."

"single pixel odd deviations" was meant to be a cheeky reference to 'noise' and hence why I put "rescue" in quotes.

I guess I just was a bit too stylish...perhaps this then:

"Cheap, 8 bit camcorders are noisy and recording their signal to a 10 bit codec may very well show you the noise whereas the 8 bit codec may mask it due to the very fact that it isn't precise enough to actually show it, so 10 bit recording of a camcorder with a relatively high noise floor may actually look worse than 8 bit."

How's that?

[Steve Kalle]

Tim,

So what are your thoughts on recording 10 vs 8 bit with the EX cameras? Is it worth it with the EX1/3 or only 350?

[Tim Kolb]

I've had mixed results with a EX1...I'd say I had obvious advantages to using 10 bit with greenscreen shots...

For everyday shooting...eh...well, let's just say I haven't come to my opinion on these things by postulating theory all day...

Disclaimer: I use a variety of external recording devices, 8 and 10 bit...they all have their sweet spot, and I use whatever is appropriate to the project.