selmslie Loc: Fernandina Beach, FL, USA

The main issue with the DXO DR plot is that it assumes that the DR at base ISO is measured from the maximum sensor value (where there is no visible noise) down to the level of the read noise (where there is only noise, no image). That's a simplistic version of DR that makes no sense since it's impossible to recover information at the bottom of that range.

The P2P results are more reasonable. The maximum DR is reduced by over two stops at base ISO for the A7 III. Then, where the dual gain processor switches, there is a slight increase in DR (about a stop) which continues in a straight line to the highest analog ISO setting. Without the second range there would have hardly been any DR at all at the end. And the line can reach a higher ISO.

In any case, both the initial range and the second range of the DR plot are perfectly straight for the A7 III. Some other cameras are not as straight and some are even a bit wobbly, even cameras without dual gain processors.

But questions remain.

1. What is the rationale for the DR reduction at base ISO?
2. What is the relationship between the increase in DR and the read noise when the processor switches to the second range?
3. Is the ending DR based on different criteria than the beginning DR at base ISO?

selmslie Loc: Fernandina Beach, FL, USA

To answer this question, here is an the illustrations from Camera Dynamic Range and Signal to Noise Ratio

At around ISO 500-640 the DR shown by P2P is about 9½ stops. The top of the DR is three stops above the level shown for EC+0 leaving about 6½ stops below for the remainder of the DR. As you can see in the attached plot and table, there is a 0.11 stop drop in SD between ISO 500 and 640 of at EC-4. It's a little bit more at EC-5 (0.86). It's safe to assume that it would be over 1 stop at EC-6. P2P indicates that it should be only increase about 0.4 stops.

But since we can only use one ISO setting at a time, the only way to judge the effect on our photos is to see how they respond to changes in brightness or exposure at a single ISO which has a single read noise level. It's not likely that we would observe a difference in recovering shadow detail by switching from ISO 500 to 640.

selmslie Loc: Fernandina Beach, FL, USA

In fact, the change in SD is only an indirect indicator of loss of DR as the gains are switched.

Each time the maximum exposure goes down by 1 stop the DR also drops by 1 stop. Note that the P2P DR is parallel to the Max Signal along each of the two gain segments. If the read noise itself were a significant contributor the DR calculation the DR would be dropping off by more than the drop in the Max Signal - they would be moving apart.

selmslie Loc: Fernandina Beach, FL, USA

They are not moving apart because the drop in DR is only related to the drop in exposure.

I collected a series of samples at EC-5 for the Z7 and got the raw values for ISO 64 through 6400. When I plotted them they were almost flat. Because I was looking at EC-5 I could only test the Z7 up to ISO 6400 but that's enough to cover the range where the gain switches from ISO 320 and 400. The Z7 also does not closely follow the exposure triangle above ISO 1600 but that is not relevant to this test.

Then I plotted the read noise values and they show how the read noise increases dramatically as the ISO increases.

Finally, I subtracted the noise from EC-5 raw values to get the DR. Because the exposure compensated for the change in ISO, the raw values remained (almost) constant in conformance with the exposure triangle. In other words, the DR does not change with ISO when the light value (LV) is constant.

This also shows that there is no evidence of a discontinuity in the DR when changing the ISO. Read noise plays no part.

I will run another test later with the A7 III at EC+0 to confirm this. It follows the exposure triangle more faithfully and the higher exposure will allow me to measure the full ISO range.

selmslie Loc: Fernandina Beach, FL, USA

You can't see a difference between the two versions of the output. They are probably more consistent is because the A7 III was set to EC+0 and the Z7 at EC-5.

Since both tests show no effect on DR on either side of the change in gain is that the read noise changes proportionately with the exposure. Consequently, the DR does not change over those spans.

But the SD does change at the transition from one gain program to the next and this changes the DR. This only happens once for each camera. The manufacturers know how much to change the final raw values to maintain compliance with the LV relationship to raw value. But the change in DR has to be deduced from the change in SD. Whether it's calculated or just an educated guess does not matter.