CineD Lab Test Discussion: UCine12K & PYXIS 6K Sensors

Here's the podcast:

Just starting to watch this, but may have to cut short to run to play golf with my mom as that was a promise for the day. But I wanted to share this with you. I'll return to the thread once I finish it myself, which may be later tonight.

Enjoy.

Interesting bits of information provided in the video on the CineD Lab Testing practices.

Some good info on the BMCC6K Sensor Results and the URSA Cine 12K. Especially with how well the UC12K performed.

What fascinated me most was the discussion on Bit Depth. I agree that it would be great for camera manufacturers to start pushing the boundaries of Bit Depth. More so than resolution. If Blackmagic could start getting 14-Bit LOG out of their sensors they may see an improvement in performance. Maybe a boost in dynamic range and exposure latitude.

Either way, a good discussion. I have a lot of good faith that Blackmagic is going to do well in 2025 with their URSA Cine line.

I don’t regularly follow any podcasts primarily because they tend to be too long saying too little as if the podcaster is paid by the minute like a journalist paid by the word count.

That said, hasn’t CineD heard of the concept of video chapters? This should have had about a dozen chapters. The URSA Cine 12K LF coverage starts at 1:14 if you want to save an hour of your life.

Agree they are favorable in their remarks about the Cine 12K but Nino questions BMD’s pursuit of high-end markets that’s “a tough sell for BMD given their history.” I wonder what they’ll say of value about the Cine 17K next annual review in 2026. “Nobody shoots 17K, eh?”

...What fascinated me most was the discussion on Bit Depth. I agree that it would be great for camera manufacturers to start pushing the boundaries of Bit Depth. More so than resolution. If Blackmagic could start getting 14-Bit LOG out of their sensors they may see an improvement in performance. Maybe a boost in dynamic range and exposure latitude. ...

At least 14-bit definitely needed to do HDR properly... yea, in my opinion.

The reason color depth has been at 10-bit is the computation times of GPUs and keeping to a single precision. Even though GPUs can do everything in 32-bit float, that takes twice as long... but maybe will be much faster with new hardware. Looking forward to getting the M4 Ultra Studio when it comes out in a few months.

Remember the cameras do 16bit linear internally in the camera and the 12bit log is converted back to 16bit in DaVinci Resolve for colour processing. I agree shooting 14bit rather than 12bit should be an improvement in colour on BMD’s high-end cinema cameras. However the extra bits might impact maximum frame rates and certainly requires additional media storage. A camera that supports 36 fps 12bit might only allow for 30 fps 14bit.

At least 14-bit definitely needed to do HDR properly...

Yeah, I mean, how do people expect to shoot anything worth watching on an Alexa35 when its flavour of ArriRaw only goes up to a measly 13-bits! Don't even get me started on how bad the 12-bit ArriRaw looks on the other Alexa bodies... ;p

Seriously though, these kinds of "lab tests" are interesting but only to a point. This forum has an infuriating habit of missing the forest for the trees when it comes to cinematography. I hate to be one of those "touch grass" people, but if folks spent as much time working on their craft as they did obsessing over numbers I think they'd discover that bit depth is not the thing standing between them and good work.

Understand, BMD cameras and Resolve make beautiful images as is, but it’s still a good thing if you could add a bit or two when capturing a gentle highlight roll off without using highlight recovery that admittedly does work well in some scenarios when its colour ‘creation’ is obvious (such as when those intense theatrical LEDs blow up in highlights). In general though I don’t use that tool as a default since it may not be kind to more complex surfaces and colours where it’s better to accept the clipping. You are right of course that one shouldn’t obsess about it.

The reason color depth has been at 10-bit is the computation times of GPUs and keeping to a single precision. Even though GPUs can do everything in 32-bit float, that takes twice as long... but maybe will be much faster with new hardware. Looking forward to getting the M4 Ultra Studio when it comes out in a few months.

The Resolve image pipeline is already 32-bit float. Everything you ingest into Resolve will be converted to 32-bit float for processing, unless you don't change anything and "Bypass re-encode when possible" is enabled (for codecs it can work with).

On acquisition the higher bit depth the better. That's partly why the Alexa 35 going to 13-Bit LOG (17-Bit Linear) was such a big deal. They certainly got a boost in the quality of color manipulation in Resolve. Whether there was a boost in dynamic range is tough to say considering they advertise 17 Stops for the camera as a whole. Whether the real boost is in ARRIRAW 17-Bit Linear / 13-Bit LOG is to be seen. ProRes nonetheless benefited from the increased sensor dynamic range.

The main thing is the ability to push colors in post grading. 12-Bit LOG is roughly 4,096 Shades of RGB (68.72 Billion Colors). 13-Bit is 8,192 Shades of RGB (549.75 Billion Colors). Taking things to 14-Bit would result in 16,384 Shades of RGB (4.34 Trillion Colors).

Yes, we've been doing well with 12-Bit. Obviously Arri has now set a new gold standard with 13-Bit. Can we really tell? I don't know.

And, as stated the tools are all good enough. The higher numbers shouldn't matter. The main thing is using the tools.

For me the reason I want Blackmagic to keep pushing is because it simply makes the tools better. If we have more dynamic range and exposure latitude then we have more room to play in post. Higher bit depth means more room to manipulate the colors in post.

If anyone could go for a higher bit depth it is Blackmagic because they have Resolve to work with it in post. They can take better advantage of it. Plus if higher bit depth is the secret to more dynamic range and exposure latitude then why not go for it?

On acquisition the higher bit depth the better… if higher bit depth is the secret to more dynamic range and exposure latitude then why not go for it?

This is the most salient point with a digital sensor’s current design. It’s not about wanting a trillion different colours; it’s about not breaking the image. A controlled shoot with sufficient light may be fine with 12bit log, but there are occasions in which images will clip when you don’t have total control of the sources of light. It’s easy to clip at least one colour channel and that’s all you need to distort the colour in post. An extra bit or two in the highlights may suffice to rescue a shot when other alternatives are not feasible or available.

Film has the ability to roll off highlights more gently as the response to light is not linear. But sensors haven’t incorporated that gentle response curve yet. So add a bit or two and it’s likely to prevent clipping other than some specular highlights which look fine clipped as our own eyes are used to specular highlights having no detail.

An alternative would be to find a formula that does mimic the behaviour of film. Now if one could develop that, there’s a patent worth protecting.

Higher bit depth is not the secret to dynamic range / latitude etc. It is merely the solution to storing a higher dynamic range into a normalized integer log encoding without quantization/precision issues that would be visible once formed back to an image in SDR or HDR. The Alexa 35 records at higher depth because it's sensor is capable of recording a dynamic range that was higher than their previous LogC3 formula could do. Hence why 10bit ProRes flavors also aren't available on that camera.

If this would be the case with a BMD sensor as well then only that would be a reason to go higher, together with possibly a new log formula.

If anyone could go for a higher bit depth it is Blackmagic because they have Resolve to work with it in post. They can take better advantage of it.

Why would Resolve matter in this context? Any NLE, DCC or grading and/or finishing software that has proper color management can take advantage of any supported log or raw format.

Film has the ability to roll off highlights more gently as the response to light is not linear. But sensors haven’t incorporated that gentle response curve yet.

And they should never do. Once a camera starts to record in non linear fashion we would be back to square one with compositing and grading. We creatively apply this roll-off in post via grading and/or by using any display transformation like DaVinci, ACES or ' camera standard' LUTs and this is typically desired to be the final step in forming the image. Definitely not the first.

...What fascinated me most was the discussion on Bit Depth. I agree that it would be great for camera manufacturers to start pushing the boundaries of Bit Depth. More so than resolution. If Blackmagic could start getting 14-Bit LOG out of their sensors they may see an improvement in performance. Maybe a boost in dynamic range and exposure latitude. ...

At least 14-bit definitely needed to do HDR properly... yea, in my opinion.

Then there’s no cinema camera that exists today doing HDR? Maybe should let all those big studios know their HDR masters are bogus.

There isn’t a 14bit LOG video camera that I know of.

Only Arri’s very recently introduced Alexa 35 does 13bit LOG. Previously was 12bit until this model.

14 bit LOG would dramatically increase the data rate.

JB

More data you say…


Sent from my iPhone using Tapatalk

It’s not just the data rate though.

It’s all the other computational overhead.

It’s a knock on effect that affects the integration time (rolling shutter) and also the maximum possible frame rates, the amount of memory required for processing etc.

Most people don’t realise how significant it is that the Ursa has so many monitor outputs, but each one added has a computational “cost”. Think about the addressable monitor outputs on Ursa Cine. That means they each need to be debayered in real time and have different overlays applied.

A common complaint with mid range cameras like Burano is the limited number of outputs. There’s a kind of end user speculation it’s a choice being made to limit a feature to protect higher end models.

I’d not be surprised that it was designed that way for similar reasons. It’s actually hard / expensive to do, deceptively so.

Making that a 14 bit workflow adds a huge knock on to all those design considerations.

14bit would be great. But I’d rather have faster rolling shutter and more monitor outputs in a smaller camera that doesn’t need as much cooling thanks.

JB

...The Resolve image pipeline is already 32-bit float. Everything you ingest into Resolve will be converted to 32-bit float for processing...

yes but... look at the limitations of the GPU and the amount of time to process... especially if trying to get real time playback.

... Maybe should let all those big studios know their HDR masters are bogus...

The key word in my opinion was "properly"... At least 14-bit definitely needed to do HDR properly. And yes much of the HDR being done is pretty good, but most looks lacking (or is just done improperly... maybe with the excuse of artistic license.)

Hey, not the CineD test, but just found this interesting look at the 12K with some additional info.

Someone else interested in it's Super16 ability.

The key word in my opinion was "properly"... At least 14-bit definitely needed to do HDR properly. And yes much of the HDR being done is pretty good, but most looks lacking (or is just done improperly... maybe with the excuse of artistic license.)

I'll beg to differ.

What is it exactly you think it's being done properly with current HDR? What is it that's lacking that having extra bit depth would manifestly change?

HDR has been a meaningful revolution in my own work and if your answer is "more dynamic range" then I don't think you understand how HDR mastering is being done today.

JB

… HDR has been a meaningful revolution in my own work...

JB

I agree with this impact within the context of the 12bit media BMD provides. My first grade from my shoots is always delivering in HDR ST2084 and that version gives me the most pleasure, then I add Rec.709 and sRGB deliverables that the clients use.

The discussion about 14bit colour comes from the speculative musings of CineD which aren’t invalid in their context of increased latitude, but may not be practical to implement in our cinema cameras given the tradeoffs that may be required today as JB has indicated.

The bit depth needed for HDR is 12 bits according to the inventor, Dolby. With 12 bits, gradients above 30 nits are imperceptible because of the diminishing sensitivity of the human visual system as brightness increases. The ST2084 EOTF can code the entire 10,000 nit range without introducing any artifacts using 12 bits. See page 9.

https://studylib.net/doc/18855695/dolby ... hite-paper

What fascinated me most was the discussion on Bit Depth. I agree that it would be great for camera manufacturers to start pushing the boundaries of Bit Depth. More so than resolution. If Blackmagic could start getting 14-Bit LOG out of their sensors they may see an improvement in performance. Maybe a boost in dynamic range and exposure latitude.

In regards to the majority of sensors, what would the point of 14b log even be if theyre using the 12b readout only? Your recorded data is not going to be better just by being in a bigger container. A 12b log container can contain up to 16b of linear data, seems like this is not the issue.

Even if the readout would be in linear float 32 initially, the integer storage container should only ever be as large as necessary for the log formula used. And the formula used is typically constrained to just enough dynamic range to encompass what the sensor can capture. Any additional storage data beyond what the sensor sees would be a waste in processing power, writing speed, storage space, and image quality.

... The ST2084 EOTF can code the entire 10,000 nit range without introducing any artifacts using 12 bits. See page 9.

https://studylib.net/doc/18855695/dolby ... hite-paper

Yes, I am aware of that paper. Note in that very graph that the visual difference crosses above the threshold at around 32 nits. Also this approach assumes that the noise (or error from calculations like rounding and others) is zero when in reality usually takes up the lowest couple bits which impacts the accuracy of dark values. And unfortunately this follows that one will have noticeable noise errors when using half precision calculation. Dolby even states that they looked for an "efficient" way... and SMPTE ST-2084 works very well, however it is not without problems for critical work at lower luminance. For example, when making an inverse of a color transform... even if not visually apparent the math shows issues (typically a result of rounding error.) In my opinion these issues will lessen significantly when using higher precision and a bit more bit depth.

Of course more bits is better.

But it’s a gigantic leap to have a camera sensor pipeline that can do it for a marginal increase in “HDR” -ness!

JB

... The ST2084 EOTF can code the entire 10,000 nit range without introducing any artifacts using 12 bits. See page 9.

https://studylib.net/doc/18855695/dolby ... hite-paper

Yes, I am aware of that paper. Note in that very graph that the visual difference crosses above the threshold at around 32 nits. Also this approach assumes that the noise (or error from calculations like rounding and others) is zero when in reality usually takes up the lowest couple bits which impacts the accuracy of dark values. And unfortunately this follows that one will have noticeable noise errors when using half precision calculation. Dolby even states that they looked for an "efficient" way... and SMPTE ST-2084 works very well, however it is not without problems for critical work at lower luminance. For example, when making an inverse of a color transform... even if not visually apparent the math shows issues (typically a result of rounding error.) In my opinion these issues will lessen significantly when using higher precision and a bit more bit depth.

Okay but the audience research behind Dolby PQ already considered what was noticeably important, thus it was called a "perceptual quantizer." If you want to be exceptional, or for that sub-32-nit region specifically; data points could be reapportioned from the existing, under utilized 12b data space above 2500 nits without resorting to 14b.

But, when you shoot BRAW, the camera doesn't know if your final output target will be HDR or regular 709. The camera thus captures plenty of low brightness information always in BM Film Gen 5, but when outputting for HDR you're not apportioning all of it across the full 12b word length of the PQ transfer if you're brightness levels are below 2500 nit, thus it's not really a problem of camera capture. The problem is the redistribution of camera data points into the PQ viewing bucket that doesn't use the full word length.

Camera log has already reapportioned linear light for the benefit of the under-served shadows during capture. Afterward, it's the PQ output for audience viewing in 12 bits that has the truncated word length unless there are data points above 2500 nits in the transfer, otherwise wasted space. But 12b PQ has been judged by Dolby's controlled viewing to be perceptually artifact free as stated in their white paper and that's the important point, not math rounding errors.