In 10 days we'll know what ISO400,000 looks like!

[BJL]

The problem with those samples is that it appears that the D4s samples have been heavily stepped-on with noise reduction while the D4 samples haven't.  The D4s samples all have zero chroma noise, and a lot of smearing, all evidence of heavy noise filtering.

Agreed … not to mention that a number of the D4s samples seem to be out of focus (or is that just massive NR?). My only tentative conclusion from them is the negative one of confirming EI=6400 as a rough upper limit for images that are usable at full resolution -- with the caveat that due to the signs of heavy NR, the actual practical limit could be lower.

[BernardLanguillier]

Agreed … not to mention that a number of the D4s samples seem to be out of focus (or is that just massive NR?). My only tentative conclusion from them is the negative one of confirming EI=6400 as a rough upper limit for images that are usable at full resolution -- with the caveat that due to the signs of heavy NR, the actual practical limit could be lower.

BJL,

What do you mean by image that "are usable at full resolution"? What output size does that correspond to?

Besides, I am still unclear about the value of debating very poorly captured samples that appear not to even have been optimally focused.

Cheers,
Bernard

[BJL]

BJL,

What do you mean by image that "are usable at full resolution"? What output size does that correspond to?

Besides, I am still unclear about the value of debating very poorly captured samples that appear not to even have been optimally focused.

Cheers,
Bernard

I agree with the focus issues (which muddy the D4 vs D4s comparisons), but that will not make the noise look worse than with properly focussed images.

Actually, I did some more (still crude) testing, and might accept up to the maximum official ISO speed of 25,600 as giving "decent" results at full 16MP resolution, albeit with the help of significant NR in those JPEG samples.

Roughly what I by mean by "usable at full resolution" is how the 16MP files look at a typical combination of print resolution and viewing distance, like 200 to 300 PPI and 15" (not the nonsense of close scrutiny at 100% on a 125PPI screen).
My two quick and dirty tests on those 100% crop JPEGs are:
1) Display at 100% pixels on my 125PPI monitor, but view from 30": that gives about "3750 pixels per viewing distance", a bit beyond the resolution limit of my less than 20/20 vision, so my eyes are dithering the individual pixels a bit.
2) Display at 50% pixels, so about 250PPI monitor, and view from 15", which is about my minimum focusing distance these days.

Under those conditions, the 100% crops make it fairly clear that by the official maximum ISO exposure index setting of 25,600, there is quite visible noise, even after the heavy NR applied to those JPEGs, and the decline at EI=51,200 and above is obvious.

But this does not bother me!  (It is due to fundamental limits of photon physics after all, not of the camera.) That 250PPI gives an image about 20" wide, and there are many cases where a smaller image and thus higher PPI is very useful with low-light photography.  At 500PPI (25% on my screen, so downsized so 1232 pixels wide) corresponding to a 10" wide print, up to EI=102400 looks decent enough for many journalistic purposes, and that 1232 pixel wide is plenty for a news website (as opposed to a pixel-peeping website).

[BJL]

The D4 has a full-well capacity of 120k e-, according to Marianne Oelund's (and others') measurements.

Thanks; do you have reference for Marianne Oelund's work? I have heard her quoted with approval a number of times.

Of course, deeper wells have no effect on low light, high EI performance; they just lower the base-ISO speed, Ssat.
But the new data do require a slight update of the above calculations. The other useful number is the DXO measurement of the base-ISO speed of the D4 as 75.  This means that at EI of 75, the image of a bra caed or mid-tones gives a signal of 12.7% of that 120,000 full well capacity, or 15,240e-. So at EI=11430, the gray card gives a signal 100e-, and the photon shot noise is sqrt(100) = 10, so the SNR is at best 10:1 (less with any noise added by the camera itself.)
Conclusion: the Snoise10 upper limit of the ISO speed latitude as defined by the ISO is at most 11430.

Pushing on to EI=25,600, the "mid-tone" electron count drops to 45 and the SNR at mid-tones to at best 7:1.
At 409600 it is about 2.8e- per photosite and mid-tone SNR at most 1.7:1.
Emil Martinec has an illustration of what SNR from 1:1 to 8:1 looks like at http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p2.html#SNR-DR

[figure1a]

Ordered mine!

[bjanes]

One slight change though: if the exposure index is calibrated as the ISO defines, the mid-tone (image from an 18% gray card) is placed at 12.75% of maximum, not 18%, to allow room for the average reflectivity of a scene being a bit less than 18%.

Your point is well taken, but one does not have to expose according to the light meter reading and leave 0.5 EV of highlight headroom. I chose the 18% figure because that is what DXO uses in their tests.

Regards,

Bill

[BJL]

Your point is well taken, but one does not have to expose according to the light meter reading and leave 0.5 EV of highlight headroom. I chose the 18% figure because that is what DXO uses in their tests.

Regards,

Bill

Actually, the traditional 18% reflectivity standard is irrelevant to either the ISO sensor saturation based measure Ssat of minimum safe exposure index or to the DXO measurements of what it [wrongly] calls ISO sensitivity.  Those measures are based on the illumination level reaching the sensor from a uniformly illuminated subject like a gray card and how much exposure is needed to send that signal to full well capacity, measures that do not know or care what combination of incident illumination and subject reflectivity are used to deliver that illumination level to the camera.

To quote DXO,

The saturation focal plane exposure Hsat is defined as the exposure (illumination multiplied by exposure time in lux.s) necessary to reach sensor saturation. ISO sensitivity is then defined by Ssat = 78/Hsat
--- http://www.dxomark.com/About/In-depth-measurements/Measurements/ISO-sensitivity

That factor of 78 is what ISO12232 specifies, and it is based on giving the exposure index that sends the uniform illumination level to 12.7% of saturation level, not 18%.

One that is understood, the DXO graph for "ISO sensitivity" show a minima value of 75 for the D4, occurring at EI setting of either 50 or 100, so 75 is the EI at which the light coming from any gray card or metered mid tone level is sent to 12.7% of full well capacity. The rest follows as in my more recent post above.


P. S. On the other hand, DXO also publishes graphs of SNR at 18% at various EI settings, but I cannot find their explanation of how those measurements are made: 18% of what?  Can you point me to an explanation of those graphs?  If one uses those "SNR 18%" graphs, one probably does get an idea of when the SNR for a subject of 18% reflectivity hits SNR thresholds like 20dB which is 10:1 as in the ISO Snoise10 measure of the upper end of the exposure speed latitude.  (Of course, to judge low light performance, one should read the actual EI as given by the camera, because that measures how much illumination the sensor is getting in the DXO tests.)  Just be aware that this DXO measurement will give a somewhat higher EI value than ISO Snoise10, by a factor of about 18/12.7.


P. P. S. the myth that a typical scene has an average reflectivity of 18% was debunked a long time ago, so there is no half stop "safety factor" or "underexposure" involved in the use of 12.7% in the ISO definition and DXO measurements.  Still, measures at 18% are useful for comparisons between cameras.

[thierrylegros396]

Deception !

Altough slightly better than the D4, it seems to be inferior to the DF and D3S for low light.

http://www.dxomark.com/Reviews/Nikon-D4s-sensor-review-Master-of-Darkness

Or go directly to the conclusion

http://www.dxomark.com/Reviews/Nikon-D4s-sensor-review-Master-of-Darkness/Conclusion

Have  Nice Day.

Thierry

[Petrus]

I tried D4s at maximum ISO, it does make a recognizable picture… Nothing much to e-mail home about (and attach files) , but I guess even a pointillistic picture is better than nothing...

[LKaven]

Bill Claff is finishing up his measurements:
http://www.dpreview.com/forums/post/53278734

[Theodoros]

Bill Claff is finishing up his measurements:
http://www.dpreview.com/forums/post/53278734

I wonder what they smoke in DXO… http://www.dxomark.com/Cameras/Compare/Side-by-side/Nikon-D4s-versus-Nikon-Df-versus-Nikon-D4___945_925_767

[bclaff]

Bill Claff is finishing up his measurements:
http://www.dpreview.com/forums/post/53278734

Done! 
( http://home.comcast.net/~NikonD70/Charts/PDR.htm#D4,D4S )

[BJL]

Done!  
( http://home.comcast.net/~NikonD70/Charts/PDR.htm#D4,D4S )

Thanks Bill!

These data and those from DXO seem to indicate the same conclusion: that improvements in extreme low light, high EI ("ISO") samples from the D4s are primarily due to better NR processing, perhaps in the production of raw files, or just in the raw-JPEG conversions used for the comparisons. (Not that better out-of-the-camera JPEGs aren't a good thing for sports/PJ usage!)

To anyone who has seen evidence of the reported improvements in low-light performance, can you comment on how the files were produced? (E.g. were they in-camera JPEG's or ...)