CineD Hit Piece on Sony Provokes Community

[jonpais]

CineD recently published an article in which the author disputes Sony’s claim that the FX9, FX6 and FX3 shoot 16-bit RAW video but brings no facts to the table, only unfounded speculation. The very first paragraph begins, “In fact, a true 16-Bit sensor readout only belongs to high-end cinema cameras. Yes, that’s one of the things you’re paying for!”

CineD starts off badly with a demonstrable falsehood. The RED KOMODO, an entry-level cinema camera that sells for just $6,000, shoots 16-bit REDCODE RAW. The author’s argument that only high-end cinema cameras shoot 16-bit RAW, therefore the FX9, FX6 and FX3 can’t possibly be 16-bit is a verifiable falsehood.

Furthermore, the author states: “In fact, the bit-depth of the RAW file is chosen based on the actual, measured dynamic range of the sensor. If a camera sensor can capture 13.2 stops of dynamic range, for example, it doesn’t make sense to use 16 bits: in this particular situation, 14 bits are more than enough!”

Here, the author asserts that “It doesn’t make sense to use 16 bits”. Seriously? Again, this could equally apply to the RED KOMODO, which measures less than 13.2 stops dynamic range according to CineD’s own lab tests. But never does CineD question RED’s claim that their cameras record 16-bit RAW. According to the author’s own logic, since the RED KOMODO is affordable and has a respectable if not awe-inspiring dynamic range, it must not record 16-bit RAW. Why does CineD not question other manufacturer’s claims as well? It should be noted that even cameras that only shoot 8 bit can cram 12 or so stops of dynamic range into their files.

In response to a reader, the author writes that Sony “refused to provide this information by disclosing the bit-depth used by the Analog-to-Digital Converter when shooting RAW. If the ADC was actually 16-bit, it would have been remarkable for this price range and Sony would have been happy to announce such a high-end feature for this cameras [sic]. Don’t you think so?” First of all, the Sony FX9 costs twice as much as the RED KOMODO, so it is not so surprising to anyone with more than two brain cells. That Sony did not respond to CineD’s request or that the author cannot fathom how a feature formerly only found in high-end cameras could possibly be implemented in prosumer models is not proof of anything. For example, Sony wouldn’t confirm or deny whether the a7s III had dual base ISO even when filmmakers showed how the sensor behaved. And if CineD contacted all the major manufacturers in the world and began asking them for proprietary information, they wouldn’t provide them with it. Perhaps most significantly, in general, camera manufacturers do not boast about ADCs in their marketing.

Finally, the author confuses feelings and assumptions with facts. “Furthermore, a true 16-Bit sensor readout, paired with backside illumination technology, also meant that this new generation of Sony cameras would provide a huge dynamic range and image quality improvement over the old FS7 and FS5 models. Or at least, that’s what we were all expecting.”

Really? How can the author presume to know the dynamic range everyone was expecting to see out of the Sonys? However much we searched, we couldn’t find any evidence for the author’s bold claim anywhere on the Internet. If anything, reviewers were very impressed with the dynamic range of the Sony cameras. Did the author truly expect cameras costing just a few thousand dollars to surpass ARRI in dynamic range? If the author finds it “very hard to believe that these sensors have a 16-Bit readout”, we find it incredible that CineD would not only publish, but defend, this sort of rubbish.

 

[cyvideo]

An article that technically had very little empirically to support the assertions made by the author. I have seen no other publication or tech analysis that would support the contentious claims made in the CineD article.

Chris Young

 

[Tom Roper]

Chris, I don't see it as a hit piece. The author states it's just his opinion and I do see logically what drives it. The perceived problem is that FX3/6/9 aren't showing greater DR than A7SIII and Alpha1. Both A7SIII and Alpha1 are spec'd with 14 bit imagers. That puts a hard limit on DR at 14 for the A series. But FX3/6/9 imagers are spec'd at "15+" stops DR. They must have a 16 bit imager to achieve that. I assume that they do but nevertheless aren't hitting a higher DR number than the A series. The FX series is apparently not bottlenecked by the A/D but something else, perhaps the sensitivity is not there or some sensitivity is being consumed by other devices in the optical path, or maybe it owes to not having the same imaging sensors as used in the A series.

DR is fundamentally a measurement of sensor range only, but what imatest attempts to do is normalize it to a level of usable range with S/N. The manufacturers are not bound by that; they are not quoting usable stops, just stops. And therein is the rub, they can manipulate the S/N factor with N/R.

Cined is doing a better job with a messy problem than anyone else. Gunther has been fairly reporting on cameras. Cined takes arrows whenever someone perceives their ox is getting gored. Gunther points out whenever there is potential to mine more DR, through NR, through highlight recovery.

Consumers are seeking one number to rate cameras. Rank is everything. At one time it was number of pixels. Now it's DR. I really pity that because they live and breathe on these numbers. It is folly.

 

[DLD]

Let me put my 2-bits in - Sony was given a chance to counter the charge. Sony punted. Charge sticks.

 

[jonpais]

Let me put my 2-bits in - Sony was given a chance to counter the charge. Sony punted. Charge sticks.

Fortunately, mankind has progressed to the point that anyone with an IQ higher than 30 can recognize such farcical reasoning for what it is - BS. Reach out to any camera manufacturer and they're not about to give you details either. Sony wouldn't even confirm that the a7s III has dual native ISO. Wouldn't that be something to boast about, too, to paraphrase the author of the hit piece over at discredited CineD? That would also imply that all of Atomos' marketing of 16-bit linear to 12-bit Log is deceptive.

The following is from an interview by Nino Leitner of former Atomos CEO Jeromy Young in 2019.

Nino Leitner: So how do you actually… you’re able to record the full 16-bit of the FX9, because we talked to Sony a week ago, they were like, “we’re talking to Atomos, but we’re not sure if there’s a recorder already in the market from them that is actually capable of doing it”.

Jeromy Young: Yeah, so that’s why the important step was at WWDC which we haven’t spoken since then and there was a bit of confusion around where the 8k ProRes RAW files came from for the Mac Pro launch and I was sitting in the front because I was part of that team that recorded those 8k ProRes RAW files and obviously they were focused on their computer a bit more than they were on where the files came from in that presentation. But behind that, we had a press release approved by Apple that we have an 8k module, we call it master control unit, for the back of the Neons. That goes to 8k 60p RAW. Now, that can do the 16-bit. And that obviously is a higher end product. People who understand what 16-bit is probably are more in the film end because it is a pretty heavy workflow even for today’s computers. And we’ll have to do it in that product because of the data rates.


NL: So you can’t use a Shogun 7 to record 16-bit RAW.

JY: No, you can’t. But we may very well be able to work with Sony on a 12-bit or a 14-bit solution which will be better than the 10-bit solution.

NL: They said it’s up to 16 bits so hopefully can do less as well.

JY: That’s right. And that’s what I’d be looking for an affordable workflow and then you can step into a higher cost workflow. And I think it’s a good solution for customers because you don’t always want to have to work in 16-bit.

NL: If you want to record 16-bit RAW from an FX9 – because a lot of people are asking this now, of course – how much would set you that Neon solution back?

JY: The good thing about the Neon is that it’s a modular system. So it’s basically a Ninja on the back and you can put an SDI module, etc., but you have to change that whole unit to go up the data rate but the 4K monitor is still relevant for even an 8k or a 6k input. So the good thing is they can use their Neon that they buy now in whatever size and I’m looking at that unit to probably be around the 5 grand mark. We haven’t announced the price yet just because we’re still mid-development and it’s high data rates. Plus, the disk support is early…

And, as has already pointed out, the very first paragraph of the hit piece begins with the claim that only high-end cinema cameras have 16-bit - ignoring the fact that the entry-level KOMODO, which costs 1/2 as much as the Sony FX9 - shoots 16-bit RAW. However, when this was pointed out to the author, rather than amend his article, he suggested that RED too might also be misleading the public, replying that "This same logic can be applied to the RED Komodo". LOL

 

[cyvideo]

Chris, I don't see it as a hit piece. The author states it's just his opinion and I do see logically what drives it. The perceived problem is that FX3/6/9 aren't showing greater DR than A7SIII and Alpha1. Both A7SIII and Alpha1 are spec'd with 14 bit imagers. That puts a hard limit on DR at 14 for the A series. But FX3/6/9 imagers are spec'd at "15+" stops DR. They must have a 16 bit imager to achieve that. I assume that they do but nevertheless aren't hitting a higher DR number than the A series. The FX series is apparently not bottlenecked by the A/D but something else, perhaps the sensitivity is not there or some sensitivity is being consumed by other devices in the optical path, or maybe it owes to not having the same imaging sensors as used in the A series.

DR is fundamentally a measurement of sensor range only, but what imatest attempts to do is normalize it to a level of usable range with S/N. The manufacturers are not bound by that; they are not quoting usable stops, just stops. And therein is the rub, they can manipulate the S/N factor with N/R.

Cined is doing a better job with a messy problem than anyone else. Gunther has been fairly reporting on cameras. Cined takes arrows whenever someone perceives their ox is getting gored. Gunther points out whenever there is potential to mine more DR, through NR, through highlight recovery.

Consumers are seeking one number to rate cameras. Rank is everything. At one time it was number of pixels. Now it's DR. I really pity that because they live and breathe on these numbers. It is folly.

Tom. I never suggested it was a 'hit piece' I think the basic mistake being made in that article is due to a misunderstanding of how most camera manufacturers get to their final output bit path descriptions. I had similar problems explaining to network buyers who were thinking of buying Panasonic broadcast camera heads for their ENG crews. Back in those days, the Panasonic WV-f700 camera heads had 8-bit signal paths on their A/D from the sensor but they were combined back through a D/A converter to derive a 10-bit signal path to the Betacam or MII recorder they were docked to. They claimed a 10-bit output. It was true the output was 10-bit. But how they got there was the question amongst engineers at the time.

So when I read a statement like the following one I see some misunderstanding of how the 16-bit linear raw output is achieved.

"When the Sony Cinema Line was first announced, I was extremely surprised to see a 16-Bit Linear RAW output on these cameras. In fact, a true 16-Bit sensor readout only belongs to high-end cinema cameras. Yes, that’s one of the things you’re paying for!"

That is a rather sweeping statement based on what? An assumption I gather from the article. Let's dive a little deeper into the articles' 16-bit understanding and how that is derived as I think, again only my "assumption" that is where the mistake is being made. For example, let's look at the ARRI Alev III sensor and see how it derives its 16-bit output.

Quote from ARRI:

"TheDual Gain Architecture simultaneously provides two separate read-out paths from each pixel with different amplification. The first path contains the regular, highly amplified signal. The second path contains a signal with lower amplification to capture the information that is clipped in the first path. Both paths feed into the camera's A/D converters, delivering a 14-bit image for each path. These images are then combined into a single 16-bit high dynamic range image. This method enhances low light performance and prevents the highlights from being clipped, thereby significantly extending the dynamic range of the image."

When Francesco states that on Gunther's findings that:

"In fact, the bit-depth of the RAW file is chosen based on the actual, measured dynamic range of the sensor. If a camera sensor can capture 13.2 stops of dynamic range, for example, it doesn’t make sense to use 16 bits: in this particular situation, 14 bits are more than enough!"

True, I believe he is right on that statement. And that is exactly what ARRI are doing for each signal path. They are delivering 14-bit images for each signal path. It's only when those images paths are combined in the second D/D stage that you derive a single 16-bit image output. At the heart of it the ARRI Alev III is processing in 14-bit on each path from the sensor, not 16-bit. Yet ARRI claims a 16-bit output. Which is a correct statement based on how they are combining their 14-bit paths. I would be highly surprised, my assumption here if Sony is not doing something very similar. Delivering 14-bit images from their sensor A/D converters and combining them into a 16-bit output. So both manufactures are valid in claiming that their output is 16-bit linear. It's just that ARRI are much more detailed or as some would say more transparent in describing how that 16-bit output path is obtained.

Chris Young

 

[Phil Rhodes]

I would point out that it's really, really easy to end up with a file that represents 16-bit numbers. Whether your entire signal path has the staggering amount of dynamic range and stunningly low noise floor to make all of those numbers in any sense meaningful is another matter entirely (that is: I suspect almost none of them do, depending greatly on gamma encoding).

As such this is really much more about what people want to put in the brochure than what they'll see on the screen.

 

[cyvideo]

As such this is really much more about what people want to put in the brochure than what they'll see on the screen.

Exactly PR.

It's the marketing fudge factor they play on because manufacturers know that the average shooter or DP is only interested in how the image looks. They really pay little heed to the pixel peepers who make up a minority of their market penetration. Referring back to those old 8-bit Panasonic heads we discovered that what Panasonic was doing was running the 8-bit output through a "hash" dithering algorithm to deliver a smooth looking 10-bit output. Much the way you may use a very light 00.01 Gaussian blur, or something to that effect to disguise banding in 8-bit graphics or 8-bit video skies.

A rather crude example of Floyd–Steinberg dithering is shown below but it will give you an idea of how manufacturers end up getting to their desired marketing number, "Oh goody it's 10-bit!" I must say though that those first-gen Panasonic and Hitachi digital heads delivered a surprisingly nice looking image when compared to the old Saticon and Plumbicon images.

Chris Young

Floyd–Steinberg dithering

 

[jonpais]

I would point out that it's really, really easy to end up with a file that represents 16-bit numbers. Whether your entire signal path has the staggering amount of dynamic range and stunningly low noise floor to make all of those numbers in any sense meaningful is another matter entirely (that is: I suspect almost none of them do, depending greatly on gamma encoding).

As such this is really much more about what people want to put in the brochure than what they'll see on the screen.

That's cute, but the reason Sony went with 16-bit RAW is that whatever happens in terms of third-party recorders - you know you have converters from 16 to 10 to 12 or even 14 - it means that their cameras will always be compatible with the highest quality, always be able to be future-proof with the 16-bit as well, so that’s also a reason why they decided to go for 16-bit rather than to a lower rate.

 

[DLD]

Fortunately, mankind has progressed to the point that anyone with an IQ higher than 30

Cine5D has been around the block and earned a solid reputation. More than Sony, which had a chance to respond but chose not to. More than a rumor site, whose job is to stimulate the discussion. And more than a jungle shaman with a penchant for a Sony brown nosing.

 

[ahalpert]

I'm tempted to ask why anyone cares about this article. But I'm probably the last person who should accuse anyone of developing a fixation

 

[jonpais]

We saw similar tripe in the online community regarding the ability of Panasonic’s S1H to output RAW over its HDMI port - a technical feat that no rational person questions nowadays:

“So, Panasonic is claiming you can record RAW video with the S1H over HDMI to a Ninja Recorder. But last I checked, HDMI is limited to 4K 4:2:2 video, so wouldn’t you just be getting 4:2:2 video in a 4:4:4 container? Same goes with the Bit depth. The S1H claims to be able to shoot 6K video in 4:2:2 10-bit, but RAW should be at least 12-bit. So wouldn’t the video still be limited to 1,024 steps vs the 4,096 offered by actual 12-bit RAW and above? (we already know that 6K is off the table with the HDMI connection.) So in summary, it seems that Panasonic and Atomos are just giving you 4:2:2, 10-bit video with a massive file size and calling it RAW”.

As it turns out, 5.9K RAW was manifestly not off the table. In fact, 8K is just around the corner. I guess it matters to those who care about the truth and are concerned about the ****-ton of misinformation being spread everywhere.