The in-camera ISO dial is a ballpark milkshake of an indicator to help choose parameters that will result in a ‘good’ perceived picture. Key ingredients to obtain a ‘good’ perceived picture are 1) ‘good’ Exposure and 2) ‘good’ in-camera or in-computer processing. It’s easier to think about them as independent processes and that comes naturally to you because you shoot raw in manual mode and you like to PP, right?
‘Good’ Exposure (controlled by shutter speed and aperture) means different things to different people but for digital cameras it means capturing the most information possible from the scene given artistic constraints (mainly DOF, acceptable sharpness/blur and desired detail in the highlights). So as long as you have maxed out shutter speed and aperture at base ISO according to your artistic intent and desirable highlights are not clipped in the raw data (sometimes referred to as exposing to the right, ETTR) you are done, right?
Well, maybe. In some cameras increasing ISO may improve SNR in the deep shadows (up to a point) at the expense of a loss in dynamic range – but as long as desirable highlights are not clipped that may mean an increase in the quality of the information captured (IQ). How far should you increase ISO? That depends on your camera. Here is how the Nikon D800 behaves in the deepest shadows (when the SNR is equal to 1 according to the relative DxOmark.com full SNR curve)
It appears that noise introduced by the electronics in the deepest shadows of the D800’s raw data keeps improving relative to the Signal as the ISO is raised from base to 800 in-camera – but from then on there is no further meaningful improvement in SNR. So if Exposure were fixed by artistic constraints and desirable highlights were still not clipped at ISO 3200 I’d limit ISO to 800 and pocket as a welcome bonus the additional two stops of DR that that setting would capture in the raw data. Nikon has a nice auto-ISO feature that allows one to do that even in manual metering mode.
The downside of this strategy is that the image straight-out-of-camera may be less bright than you are used to. But that’s not a problem for you because you want the best IQ possible so you prefer to capture a cleaner image in the raw data that can easily be made appropriately bright in conversion or PP – than to have an appropriately bright but noisier SOOC file.
Other cameras’ curves show continued shadow SNR improvement much beyond ISO 800 – for instance the Nikon D4s and most Canon DSLRs – so it is beneficial in those cases to keep increasing ISO until just before the first desirable highlights start to clip (or the maximum shadow improvement ISO is reached).
Why does the noise in the deepest shadows show this behavior? Recall that the in-camera ISO setting affects analog and/or digital parameters in the sensor, including signal gain. The curve above shows the interaction between the dynamic range of the pixel versus that of the ADC and related circuitry**.
If the pixel has a lower noise floor than the ADC, it makes sense to amplify its output analogically so that the ADC’s noise floor does not overwhelm shadow signal information rolling off the pixels. This is the case up to ISO 800 in the graph above. If on the other hand the pixel is the limiting factor in terms of noise floor, then there is no analog amplification of the output of the sensor that will improve the SNR of shadow data coming out of the ADC – and if you amplify anyways (analogically or digitally) all you are doing is reducing the capture’s DR by clipping highlights sooner. This is the case above ISO 800 above.
Some cameras like the D7000 have a virtually flat shadow noise improvement curve throughout, meaning that you do not get any benefit increasing ISO above base, you simply lose DR – so if you have one don’t do it. Unless you want a ‘good’ perceived jpeg SOOC, in which case watch your highlights. But you don’t because you want the best IQ possible and like to PP, remember?
Once you have captured the best scene information possible (best IQ), ‘good’ perceived pictures are obtained through in-camera or in-computer PP.
If you’ve read this far and you are a landscape photographer you will probably be interested in this slightly different take on the same subject.
** This is a simplification and applies to most cameras that came out until 2014. More recently Sony has started implementing Aptina’s (now ON Semiconductor’s) DR-Pix technology in their sensors, which effectively physically reconfigures the pixel itself when in-camera ISO is raised, as seen in the a7S and a7RII for instance.