I now have Read Noise values for the D800 measured from NEF files for a limited set of ISO values.

See FX Read Noise Comparison and D800 Read Noise for details.

So far, the most surprising thing is how early the D800 switches from analog gain to digital gain.

Unfortunately I have no data so far below ISO 400 so performance in that area is entirely speculative.

When we pull some gain and Full Well Capacity (FWC) out of the files; we will know more.


Bill

Visit me at My site

Links are not working for me....

Visit my Nikonians gallery.

Christopher,

Should be fixed now.

Thanks for the "heads up!"

Regards,
Bill

Visit me at My site

I'm returning to this thread to reply to a queries posed elsewhere rather than take that thread off topic.

That query (in magenta) was:
I have looked at the D800 data on your website. I am curious as to how you measured the Noise from the NEF.

Read noise was measured using the optical black data in an area of the NEF that is not used to make the image.
For some technical background you could look at Read Noise – Optical Black at my site.

As you know, the size of the D700 Sensor pixel is almost 2x the size of the D800 Sensor Pixel . S/N suggest to me that with the larger D700 Sensor (8.45 micron) vs the 4.8 micron D800 Sensor, you get twice as much light for the D700 compared with the D800 Sensor.

The Question is how does the Noise for these two different sensors compare?
I know you have done some Read Noise calculations for the D800 at different ISO's . It would be very useful once the D800 is at hand to do a like to like comparison of Read Noise for the D800/D700. Then one can get some good idea on a comparison of the S/N for the D700/D800.

However the S/N is also dependent on the Shot Noise, the Dark Noise as well as the Read Noise. These are very difficult to measure independently .I think to get a real quantitative comparison , one would have to calculate like images from a D700/D800 using the same lenses, the same settings everything and compare the raw NEF files from both cameras.
Gut feel tells me that unless Nikon has put in some extraordinary noise reduction technology in the D800, if like for like, a raw NEF D800 at higher ISOs will display lower S/N to a similar NEF from a D700 for the same settings, that is the D700 will have a better S/N at higher ISOs.
What is interesting , Nikon does not seem to publish Noise Sensor Data or S/N comparison for their different sensors. If they do, let me know where I can find it.

Some have suggested under sampling a D800 raw NEF to 12 MB, but all that does is just apply an averaging or a smoothing function over the image.

You are basically asking to compare the D800 to other cameras on a Signal to Noise Ratio (SNR) basis without regard for the size of the photosite.

The best comparison for that purpose is the Photographic Dynamic Range (PDR).
You can see a PDR Chart at my site that includes estimates for the D4 and preliminary data for the D800.

Note that technically, PDR is largely a function of read noise and gain; and that gain is interchangeable with Full Well Capacity (FWC).

Additional questions are welcome but please look at my site to see if you answer is already there!


Bill

Visit me at My site

Bill,

>Additional questions are welcome but please look at my site to see if you answer is already there!<

I didn't see my questions asked by others so, here goes:

1. On your Photographic Dynamic Range chart, preliminary results for the D800 show it to possess approximately 1 stop less DR than the "ideal maximum" value for an FX camera body. Can one correctly extrapolate that FX cannot improve beyond that limit?

2. Toward the low ISO end of the chart the results for most cameras becomes nonlinear. Why is that?

3. Related to Q2 above, is DR likely to become linear at low ISO settings in future?

Rob
www.robbuckle.co.uk

Rob,

Great questions.

1. On your Photographic Dynamic Range chart, preliminary results for the D800 show it to possess approximately 1 stop less DR than the "ideal maximum" value for an FX camera body. Can one correctly extrapolate that FX cannot improve beyond that limit?

No. I expect sensor performance to continue to improve toward the "ideal" for all sensor sizes.
But, you can see we're already getting pretty close, especially at high ISOs.
So in the future I see more people considering going to a larger format rather than a better camera at the same format if they need better high ISO performance.

2. Toward the low ISO end of the chart the results for most cameras becomes nonlinear. Why is that?

In simple terms, there are two sources of noise, one is fixed and the other varies with ISO.
As you lower ISO the fixed part remains even as the variable part gets smaller and smaller.
This effect is accentuated in CMOS designs with multiple amplification stages as opposed to CCD designs that often have only one stage.

3. Related to Q2 above, is DR likely to become linear at low ISO settings in future?

I think the "shoulder" at low ISOs will always be there although it might become of no practical importance in the future.

You could also eliminate the shoulder by sacrificing some high ISO performance.
That might be done in order to simplify the sensor design and lower costs.

Regards,
Bill

Visit me at My site

Bill,

>I expect sensor performance to continue to improve toward the "ideal" for all sensor sizes. But, you can see we're already getting pretty close, especially at high ISOs.<

Yes, I had noticed the tendency for the results to coverge with the relevant "ideal" upperbound line at higher ISO values.

Do you consider that the "ideal" lines will become increasingly hard to approach, or instead that in 1, perhaps 2 further sensor iterations we will have reached that limit?

>So in the future I see more people considering going to a larger format rather than a better camera at the same format if they need better high ISO performance.<

Is it not more correct to say "...if they need better performance at any ISO setting"?

>This effect is accentuated in CMOS designs with multiple amplification stages as opposed to CCD designs that often have only one stage.<

My understanding is that digital Medium Format cameras, with their larger CCD sensors, are usually poor performers at even moderate ISO settings compared to a smaller format sensor like FX. Yet, if I understand correctly what you have written, the slope of the DR results for a MF camera would be less than it is for a smaller format camera. That can't be correct can it?

>I think the "shoulder" at low ISOs will always be there although it might become of no practical importance in the future. You could also eliminate the shoulder by sacrificing some high ISO performance.<

Now that is interesting. The reason I asked the 3rd question in my previous post was because I figured that if the nonlinear portion is extrapolated backward from the linear portion then greater DR would be achievable at low ISO settings. You have stated that by sacrificing high ISO performance, the shoulder could be eliminated. By that did you mean made linear? If so, the obvious question to me is: for camera bodies optimised for high resolution, why isn't it? Consider, for example, the photographer who always shoots on a tripod or in a studio (or outdoors) working with strobe lighting. These photographers have next to no requirement for anything other than base ISO. The loss of high ISO performance in exchange for an increase in DR would be very welcome.

Rob
www.robbuckle.co.uk

Hi Bill
It is uncertain if the method that you are using (using Optical Black Data) can yield the information that you want (Read Noise and other Noise Data and overall S/N Ratio). It is better to ask the manufacturer (Nikon) exactly what this data is supposed to represent. The fact that it is outputted to a file does not mean that there has been some intermediate processing that one may not be aware of. Have you sent your analysis to Nikon for comment. I think it is important to do that.

There has been various attempt by different people to get to grips with the Noise Data and S/N. One simple approach was to measure the Pixel VAlue and standard deviation of a Grey Scale Photo taken at different ISO's.
I am currently evaluating all the different approaches that have been published. At least if they are consistent with each other then the S/N data should be reasonably close within the margin of error. Depending on one approach may be somewhat problematic as you may have data that is offset for some reason known only to the manufacturer.

Manufacturers dont like to disclose this kind of data (Noise Data) so I am not sure that they will oblige by writing it out as Optical Black Data for anyone to access. This information is Commercial In Confidence, its the kind of stuff which is kept in closed cabinets at some research centers. If not, then you would see all the manufacturers publishing Noise data and S/N for the Imaging Chips and fr their Cameras. The fact that they dont tells its own story.
However I believe that the information (Noise Data) is hidden in the Image and there must be a sound technical approach for extracting this data..

TIA

Steve,

I have both "optical black" and "normal" read noise data for a number of cameras and they are in agreement.
So I'm confident that the "optical black" data is correct.

There is no point in attempting to contact Nikon. They never disclose this type of information.

They could choose to exclude it from the raw data but either have not bothered or are actually using it themselves in their software.

Regards,
Bill

Visit me at My site

I cant even get a decent email address from Nikon. I have written to the only email address I know, and so far the response has not been very encouraging.

Is the Optical Black data only provided by Nikon or do other manufacturers output this type of data?

I have to look at your program, what do you do just take an average and the St Dev of the data?
You say you dont have the Gain /full well data or it is not encoded in the file. Is that correct?
TIA

Steve,

Yes, I take the average and standard deviation of the optical black values.
But, I first separate them by channel and then recombine using quadrature.
This effectively eliminates multi-channel pattern noise.

Gain and FWC are not stored; they need to be computed in the usual way.

Regards,
Bill

Visit me at My site

Hi everyone,

What every written here are very technical. I have a very basic issue and I am going no where where Nikon support. I have lots of noise big dots on everywhere on the image even at ISO800 or lower. The Nikon Service person said he does not see any issue. I hope other D800 owner's may help me whether it is normal or not. I cannot upload the image. Anybody is willing to help me, I will glad to send the image over.


Thank you,