To explain further, when you're shooting above base ISO and you blow the highlights, it's not the sensor that clips first, it's the ADC. Say base ISO is ISO100. According to Sensorgen.info's analysis of the DxOMark figures, the Full Well Capacity (FWC - the max. number of electrons that an individual sensel can hold before being completely full; any more and it's over-saturated) at actual ISO70 (manufacturer's ISO80) is 47159 e− and at actual ISO91 (manufacturers ISO100) it's 36512 e−. I'll use the later FWC as it makes doubling manufacturers ISO easier. Manufacturer ISO quoted from now on.
Sensorgen.info data for Pentax K-5
Say the light level is too low at your current aperture & shutter speed to get the histogram fully to the right. You increase the ISO from 100 to 200. Now you are amplifying the electronic output of the sensels by 2x before it proceeds to the ADC input. At base ISO, the gain between the sensor and the ADC should be set so that FWC occurs just before or at the same time as the ADC reaches Full Scale. But once we start amplifying the output of the sensels as we boost ISO, the real FWC will not be reached before the ADC clips, so this ADC clipping becomes the highlight clipping determiner, not the FWC. But if you were to consider the FWC instead, the apparent FWC (or saturation limit) drops by 1 stop for each stop of ISO boost (from Sensorgen.info):
ISO | Apparent FWC |
100 | 36512 |
200 | 18087 |
400 | 9295 |
800 | 4596 |
1600 | 2307 |
So the highlight clipping points drops as the ISO (and amplification) goes up.
The FWC is half of the Dynamic Range. The other half is the noise floor. The noise floor changes as ISO varies because some of the read noise occurs before the ISO amplification stage, and some occurs after it. (An "input-referenced" read noise figure adjusts the value so it is equivalent to electrons from the sensor.) So the DR, the ratio between the apparent FWC and the noise floor, varies in a more complex manner then just "1 stop ISO increase = 1 stop less DR".
Thus armed, let's look at the actual DR figures from Sensorgen.info:
ISO | Apparent FWC | Read Noise | DR |
100 | 36512 | 3.5 | 13.4 |
200 | 18087 | 3.1 | 12.5 |
400 | 9295 | 2.6 | 11.8 |
800 | 4596 | 2.4 | 10.9 |
1600 | 2307 | 1.9 | 10.2 |
You can see above how, due to the drop in the input-referred total read noise at higher ISOs, that the drop in DR, as ISO increases by one stop, is less than 1 stop.
These figures from Sensorgen.info use data derived from the Full SNR DxOMark values, so they're sensel-level figures. To change this to the "Print" (normalised-to-a-standardised-8MP-output-size) DR value, add 0.5 stops.
For worked examples of a Nikon D3 & K-5 that explain why total read noise changes with ISO see this post:
https://www.pentaxforums.com/forums/pentax-k-5-forum/135603-isoless-sensor-4.html#post1456059
OK, now let's look at a situation where, to get the histogram fully to the right, you boost the ISO from 100 to 400. Comparing the swings & roundabouts balance sheet from shooting with the same aperture & shutter speed at this light level at ISO100 (hence 2 stops "underexposed" according to the histogram) vs shooting at ISO400.
DR of shooting at ISO100: 13.4 stops
DR of shooting at ISO400: 11.8 stops
DR of shooting at ISO100 and boosting +2EV in PP to reach the equiv. exposure of ISO400: 13.4 - 2 = 11.4 stops (a straight-forward DR reduction)
Clipping headroom preserved by this method: (36512 e−/9295 e−) = 2 stops
Read noise increased by using this method = log2(3.5 e−/2.6 e−) = 0.4 stops
DR loss if ISO was changed instead: 11.8 - 13.4 = - 1.6 stops
So, if I've done this correctly, this shows that the DR will end up 0.4 stops smaller, using the ISO100 + 2 EV in PP method, due to 0.4 stops more total read noise, but as compensation, 2 stops of highlight headroom are preserved.
Bringing this back to the EVcomp discussion, using -1EV on the K-5 in TAv mode is not going to have much of a noise or DR impact, but will benefit the highlights. The degree of noise & DR loss/gain will depend on the ISO used in the shot.
Dan.