I’ve refined the calculations behind the graphs in the first post of this series. Specifically, I’ve added normalization to the 1.5 dB/octave lines so that they are corrected for the actual exposure of the test images, as measured by the mean value of the pixels in the selection box. The Poisson-limited lines had already been so corrected. This calibrates out exposure variations caused by inaccuracies in the shutter and the aperture, gain of the amplifiers in the camera, and variations in monitor brightness caused by power line variation and the phase of the moon.
As before, the way to use these curves is to find the ISO/DIN setting beyond which the Actual (red) line drops as faster (or faster — look at the M9 graph) as the 1.5 decibel/octave line. Beyond that point, for a given shutter speed and f-stop, you can get just as good SNR (or better, in the case of the M9) by increasing the Exposure control in Lightroom as you can by further increasing the ISO setting on your camera, and you’ll have more headroom to boot.
Now we can see that the small improvements in the Nikon D4 results at high ISOs were illusionary, an artifact of the high-ISO frames getting a bit more exposure and/or gain than the low-ISO exposures. We can also see clearly that the ISO setting below 100 is useless for reducing noise, since it just doubles the exposure.
The D800E results are substantially the same as before, but the corrected green line shows that the ISO 50 setting just increases the exposure:
The M9 results are also similar to what they were before. The camera’s ISO 80 setting is seen to operate similarly to the Nikons’. By the way, note that ISOs over 640 (DIN 29), not only don’t yield any improvement in SNR over push development in Lightroom, they actually give worse SNRs: