With the self-heating tests out of the way, I could start to see the way that the noise floor varied with ISO setting in the D4. I set the shutter to 1/30 and made 16 exposures at ISO settings of 100 through 6400, selected 90% of the image area, and averages the statistics from each set of 16 images. The average noise floor was as follows:
And the standard deviation of the noise floor looked like this:
I think the plot to concentrate on is the standard deviation curve. If the noise floor were constant, we could subtract it out. When we look at these numbers, we should also remember the long tail of the noise floor variation, which is likely to make the image look worse than we’d expect if the noise were Gaussian. In any event, it appears that the noise floor is not hugely adversely affected as the ISO is raised from 100 to 400 (or maybe 800), but after that, we lose almost a stop of dynamic range for every stop increase in the ISO setting.