NEW DATA Is the D750 still the low ISO champion???

[Frank Fremerey]

@JKoerner: Look at the astonishing career of Jim Lovelock in the atmosphere science department. He stood alone, he held his ground and he proved text books wrong. The Montreal protocol of 1984 reflected his success. Similar things seem to happen today in the gravitational science community. Only time will tell if the granting of the Nobel price was one big mistake or spot on.

So: I understand your train of thought on common wisdom and plausibility, but proof that is not.

[JKoerner007]

@JKoerner: Look at the astonishing career of Jim Lovelock in the atmosphere science department. He stood alone, he held his ground and he proved text books wrong. The Montreal protocol of 1984 reflected his success. Similar things seem to happen today in the gravitational science community. Only time will tell if the granting of the Nobel price was one big mistake or spot on.

So: I understand your train of thought on common wisdom and plausibility, but proof that is not.

There are always exciting exceptions, true.

But Nikon itself having the power to both design, and produce, all the cameras in question ... ranking/placing/pricing them accordingly ... is a lot more powerful evidence than the ability to design a data site.

As I mentioned in the first page, it all boils down what we choose to believe: the mountain of evidence or the molehill.

In civil law, all one needs to prove his case is "a preponderance of evidence" (51%).

In criminal law, all one needs to prove his case is, "beyond a reasonable doubt."

While "it is possible" everyone else is wrong, and PTP is correct, in no evidentiary system is there a requirement to prove "beyond a possible doubt."

The "preponderance of evidence" and the "reasonable conclusion" both suggest that Nikon designed/placed the D750 to be a solid, middle-ground FX camera, but not the Champion of either base or high ISO capability.

Every other testing site seems to concur, except Bill's.

[bclaff]

What kind of a target is used for PDR determination? DXO use ND filters in their noise measurements to ensure homogeneity.

https://www.dxomark.com/About/In-depth-measurements/DxOMark-testing-protocols/Noise-dynamic-range

They define the dynamic range to be the difference between maximum luminosity and the luminosity that gives SNR = 1. I would think it's exceedingly difficult to achieve accurate characterisation of noise in the darkest shadows without a highly controlled measurement setup.

Targets and measurement conditions could easily result in different results.

Photographic Dynamic Range (PDR) uses a 77-step wedge.
The densities are not calibrated but this is not relevant because we're only taking Signal to Noise Ratios (SNRs) where SIgnal is measured in Digital Numbers (DNs).

FWIW, DxOMark uses read noise which is not SNR = 1 but actually SNR = 0; the Noise when Signal is zero.
There are multiple ways to determine read noise and DxOMark doesn't actually disclose their method.
However, DxOMark Landscape scores cross-check quite well with independent read noise measurements like those at PhotonsToPhotos.

D810 and D850 don't quite match the expected DR at ISO 64 (coming down from high ISO the DR improvement slows down at the lowest ISO); my guess is that the 14-bit ADC might not be good enough to get the expected improvement at such low ISO setting.

No, 14-bits is sufficient for those cameras.
Note in the attached Read Noise chart that the y-axis is logarithmic. Read noise for both cameras is about 1.2 14-bit DN.
You don't need additional bits until this drops below about 0.7DN or so.
(Note dual conversion gain kicking in at ISO 400 for the D850.)

The D7200 could be simply a better implementation as its dynamic range follows a more linear curve as a function of ISO.  Maybe the larger sensor generates more heat?

No. No reasonable normalized measure would put a camera with a significantly smaller sensor (DX) ahead of so many cameras with larger sensors (FX).
Ultimately the area of the sensor is the most significant factor.
The underlying problem is that DxOMark made a poor definition for the Landscape score.

[bclaff]

Absolutely agree.

There is a difference between 'claims' and facts.

One method-user 'claims' his methods are better than others doesn't necessarily make this a fact.

There are good reasons, which I could detail (and have elsewhere); but I suspect you would ignore them because you can't rationalize the results.

...
As far as I know, only DxO and PTP offer graphs, which are similar, with the extreme figures at Base ISO being a bit different.
...
However, most compelling to me is the fact Nikon pushed its own D810 (and, now, D850) forward as Nikon's own champion.
...
Nikon also positions its own D750 (which, by every other measure, besides PTP) ranks second, behind the D5 in high ISO, behind the D810 in base ISO, but is a good "middle ground" between the two (better high ISO than the D810, better base ISO than the D5) but beating none of them in the extremes. The D750 is priced/positioned/spec'd as a very good mid-FX Nikon.

These facts, prices, and commercial placement by Nikon are not 'by accident'; they are by design.

With everyone else echoing this sentiment, and with Nikon itself (in both its price and placement of the D750), I just find it difficult to accept "the one" guy standing alone, offering a graph which shows the D750 offers better base ISO performance than Nikon's own flagship, Centennial champion.

My own experience, as an investigator for both civil and criminal litigation (but, admittedly, not an engineer) defaults back on what the proverbial "reasonable man" would conclude.

IMO, it is not reasonable to believe everyone else's conclusion is incorrect, except Bill's, including the multi-billion dollar Nikon Corporation, who makes all of these items ... and markets/positions each product according to their capabilities.

In the end, I do not believe Nikon 'accidentally' made the D750 with a better base ISO rank than the D810/D850, nor did they do so on purpose.

Based on all other evidence, I believe Bill's graph/placement is somehow flawed.

I have not seen any claim by Nikon that the dynamic range of the D850 is the best that it has to offer.
So the way I see it we're simply debating whether the DxOMark Landscape score of the PhotonsToPhotos Photographic Dynamic Range (PDR) tells the "true" story.
I claim PDR is the superior measurement.

As I mention above there are sound technical reasons that PDR is a better measure.
They may be beyond the scope of a simple NikonGear post (and have been discussed at length at places like dpreview).
However, if this interests you then you must at least have a good grasp of the Photon Transfer Curve (PTC).

The inability to rationalize the results ("common man") does not invalidate them.

[JKoerner007]

There are good reasons, which I could detail (and have elsewhere); but I suspect you would ignore them because you can't rationalize the results.

I don't necessarily ignore what you say ... I merely remain unconvinced by 'one guy' in the face of many saying the opposite.

I very much respect your time, effort, expertise, and passion for categorizing and attempting to 'make sense' out of all the different offerings ... I am just skeptical when your results differ from "everyone else's," including the manufacturer.

I have not seen any claim by Nikon that the dynamic range of the D850 is the best that it has to offer.

Then perhaps you should take a closer look. On the D810's product page, Nikon directly says,

• "The D810 truly raises the bar for image quality and dynamic range. An all-new FX-format full-frame image sensor design—36.3-megapixels with no optical low-pass filter—is paired with Nikon's innovative EXPEED 4 image processing for flawless detail retention from snow white to pitch black, beautiful noise-free images from ISO 64 to ISO 12,800, an extremely wide dynamic range, flattering well-saturated skin tones and much more."

There is not a single mention of 'dynamic range' on the D750's product page (which is odd, if your graph is correct, and it holds the highest spot).
This flagrant omission by Nikon of any DR reference on the D750 page only makes sense if your graph is in err ... and everyone else is correct ... namely that the D750 is not the DR champ.

Meanwhile, the D850's page references the D810, as the previous benchmark, and indicates the D850 moves beyond in dynamic range (and other attributes):

• When Nikon introduced the D800 and D800E, it set a new benchmark for DSLR image quality and super high resolution photography that approached medium format. Now, five years later, Nikon proudly introduces the next evolution in high resolution DSLRs, a camera that allows photographers to capture fast action in 45.7 megapixels of brilliant resolution. With remarkable advancements across the board—sensor design, autofocus, dynamic range, sensitivity,
  
  lower noise, wider dynamic range, subtle tonal and textural details, high-speed continuous shooting at approx. 9 fps

So the way I see it we're simply debating whether the DxOMark Landscape score of the PhotonsToPhotos Photographic Dynamic Range (PDR) tells the "true" story.
I claim PDR is the superior measurement.

We're debating a lot more than you versus DxO; we're debating you versus DxO ... Nikon Corporation ... and virtually every other expert.

Your 'claim' of PDR being the superior measurement is not necessarily a fact.

Further, Nikon engineers have more resources and means with which to avail themselves (and ascertain/develop their own sensors) than any one man ... 'common' ... or with a degree. Nikon Corporation is filled with cutting edge scientists who have access to data and measuring devises you likely have never seen.

I simply don't believe Nikon has been missing a key fact about its own camera; rather, I simply believe your graph does not represent any fact, but is merely a mistake.

As I mention above there are sound technical reasons that PDR is a better measure.
They may be beyond the scope of a simple NikonGear post (and have been discussed at length at places like dpreview).
However, if this interests you then you must at least have a good grasp of the Photon Transfer Curve (PTC).

I humbly concede my technical expertise does not match your own (at least with respect to photons). However, one of the aspects of my own profession (of ~30 years) is to hire and vet "technical experts" for courtroom appearances. Accident recon experts, who measure everything from coefficients of friction, to velocity/impact, the integrity of steel/concrete, you name it. Degreed experts are hireable commodities ... and there is much disagreement, even among 'experts' ... and not all of them are going to be correct.

Celebrated PhDs have made mistakes, decorated surgeons have amputated the wrong legs, and anesthesiologists have killed patients with mis-calculated dosages. One thing I have learned is no one is above error.

Even the most distinguished experts are subject to cross-examination, can be 'bought,' sometimes don't have all the facts, and (in short) all have the potential to suffer the same 'room for doubt' as any other witness called upon to be put under the microscope of scrutiny.

Essentially, the sum and substance from your divergence from DxO, Nikon, and everyone else is, "(You) claim PDR (your method) is the superior measurement."

While it's true I don't have the technical expertise, in your particular field, to debate with you ... I do have a mountain of expertise in assessing evidence, vetting credibility, and determining witness strength (or weakness), including 'expert' witness testimony, by looking at all the evidence.

In this thread alone, we've witnessed several mistakes already (misreading, not really checking Nikon's statements, etc.)

Hey, we all make mistakes, it's not a big deal. (Although sometimes it can be ...)

IMO, ignoring a mountain of evidence in support of 'one man's position' isn't the way to bet.

The inability to rationalize the results ("common man") does not invalidate them.

It's not my inability to rationalize; it's my unwillingness to accept 'you' as the authority over all others, including the multi-billion dollar corporation who makes all the cameras in question.

I find the statistical probability of there being some mistake in your calculations as being higher, and more likely, than every other entity being in err in their collective findings.

My last $0.02

[Bjørn Rørslett]

Please, this approach to discussion is counter-productive.

[bclaff]

Please, this approach to discussion is counter-productive.

If I may ask; to whom is this addressed?

[Bjørn Rørslett]

Sorry, should have made that point clearer: I responded to JKoerner.

Disagreement is welcome, personal attacks are not.

[Ilkka Nissilä]

Dxomark define dynamic range like this:

”3. Dynamic Range
Dynamic range is defined as the ratio between the highest and lowest gray luminance a sensor can capture. However, the lowest gray luminance makes sense only if it is not drowned by noise, thus this lower boundary is defined as the gray luminance for which the SNR is larger than 1. The dynamic range is a ratio of gray luminance; it has no defined unit per se, but it can be expressed in Ev, or f-stops.”

[bclaff]

Dxomark define dynamic range like this:

”3. Dynamic Range
Dynamic range is defined as the ratio between the highest and lowest gray luminance a sensor can capture. However, the lowest gray luminance makes sense only if it is not drowned by noise, thus this lower boundary is defined as the gray luminance for which the SNR is larger than 1. The dynamic range is a ratio of gray luminance; it has no defined unit per se, but it can be expressed in Ev, or f-stops.”

Sure, that's a common way of putting it.
But at SNR = 1 noise is very slightly higher than read noise because of the photon noise contributed by the signal.
So strictly speaking noise at SNR = 1 is not read noise; read noise is noise at SNR = 0.
I admit I'm being picky and we shouldn't labor the point. In practice the difference is not measurable.

BTW, the DxOMark reasoning (very common) also implies that we can't measure a signal that is lower than read noise; but we can.

[Frank Fremerey]

So strictly speaking noise at SNR = 1 is not read noise; read noise is noise at SNR = 0.

In "common man's language" a signal-to-noise-ratio is lowest if signal cannot be distinguished from noise. Because SNR it is a "ratio" and both signal and noise are the same at the non distiction point, my mathematical reasoning would be:

same divided by same equals one

So SNR=1 is the point where signal is undistinguishable from noise.

*

@JKoerner: Right or Wrong is not opinion for people in Science.

Law is about one Opinion prevailing over another, be it by the better argument or any other force like managing formalities and external conditions... Right, Wrong or Justice are only of rhethoric interest.

If we talk about measuring methods we talk about science, not opinion.

[JKoerner007]

If we talk about measuring methods we talk about science, not opinion.

@Frank;

Wrong: law is about the sum total of factual evidence of one side prevailing over that which can be vetted/produced by another.

"Right" or "wrong" are moral considerations ... while "correct" or "incorrect" factual ones.

In science, if we are considering the veracity of "measurements," then we also need to scrutinize the tools used to measure ... the samples measured with the tools ... and the expertise of those running the test.

A peripheral expert, with so-so tools, and one sample, will not necessarily derive the same conclusions as a lifetime expert, with the best possible funding/tools available, multiple samplings measured ... to speak nothing of the engineering expertise to design all samples considered.


@Bill;

Thanks for reaching out by PM ... and smiling at your "expert witness" appearance in court

Please don't take anything I have said as a personal attack.  I have only expressed healthy skepticism/scrutiny.

I do have your site bookmarked, and I very much appreciate your contributions, as I have tried to preface my agnosticism in certain instances

Cheers,

[bclaff]

In "common man's language" a signal-to-noise-ratio is lowest if signal cannot be distinguished from noise. Because SNR it is a "ratio" and both signal and noise are the same at the non distinction point, my mathematical reasoning would be:
same divided by same equals one
So SNR=1 is the point where signal is indistinguishable from noise.
...

I understand your reasoning and it's a very common misconception ("common man" language versus technical distinction).
The fact is you can measure a signal that is below read noise; so it is distinguishable.
In any case, any non-zero signal will have photon noise in addition to read noise therefore noise at SNR = 1 is greater than read noise.
However, as I said, in practice it is of no consequence.

[bclaff]

@Bill;
Thanks for reaching out by PM ... and smiling at your "expert witness" appearance in court
Please don't take anything I have said as a personal attack. I have only expressed healthy skepticism/scrutiny.
I do have your site bookmarked, and I very much appreciate your contributions, as I have tried to preface my agnosticism in certain instances
Cheers,

I don't feel attacked but do understand the desire for decorum which is why I took our conversation offline.
I would like to risk adding (not intending to cause a new disagreement)  that the preponderance of evidence may apply in legal matters but it's not compelling in a scientific discussion.

[OCD]

Hoo boy.  This is really getting down to some pretty fine details.  In reviewing the differences between the D850 and D750 at DxO.com the differences seem pretty minute.  DxO displays the data two ways: screen (100% zoom analysis) and print (8mp 300 dpi 8x12 print analysis).  The screen results very slightly favor the D750, and the print results very slightly favor the D850.  Honestly, when looking at these results it seems to me that not much has changed by way of sensor performance since 2012 when the D600 and D800 were released, and when selecting the D600 and D800 on DxO they perform similar to the D750 and D810 in relation to the D850.  All these sensors make great images, and the small differences on these line charts seem unlikely to manifest as any noticeable difference probably 99% of the time.  But hey, I understand if that 1% matters to you, that might be the zone you really need.  In an infinitesimal way the D850 is the best sensor built to date, heck...it's the newest.  The D750 in my estimation "competes well" because it has larger pixels, although I have no clue why the D5 does not have similar dynamic range results at the lower ISO's.  At this point we're pretty fortunate as we can choose the camera that has the features and handling we enjoy since we don't really need to worry about the sensor quality anymore.  In my estimation, the advantages of the D850 reside in all the features it is loaded with more than anything else, but sure, that 46mp sensor is nothing to sneeze at.

One area on the DxO measurements where the D850 has a more significant advantage over the D750 is the Color Sensitivity at low ISO's for Print.  Everything else seems very close to my eyes.  I think Frank mentioned something about the color rendition of the D850 being superior to anything he's seen, and the DxO test on color sensitivity would support that statement.

 

And to be honest, I am reticent to submit this post as I have only used a D750 and D7100/D5300.  I'm just looking at line charts and going from that.  But the charts do seem consistent in methodology over time, and that's about the best that can be hoped for in this case.  None of these line charts reflect what it must be like to get out and actually shoot with the camera, that's for sure.  And how we enjoy a camera is a personal and subjective decision, I try to go for what is comfortable and fun, and well....affordable for me, so I'm grateful for the quality of the sensors across the Nikon FX camera line, very much so.