Does it make sense to "Expose to the Right" except at base ISO?

Question

I have read in a number of places that exposing to the right (ETTR) only makes sense at your camera's base ISO, with the justification being that if you are boosting the gain in camera (by raising the ISO), you are negating the benefits of ETTR.

My own experience is very different. I have found that if you ETTR maximally, even at ISO 1600 (on a 5D Mk3), the noise is fairly negligible once the photo has been adjusted correctly.

Obviously the huge benefit to raising the ISO is it gives you much more room to manoeuvre, allowing you to maximally ETTR in situations where it would be impossible at base ISO.

So does it make sense to expose to the right at high(er) ISOs, or to select high(er) ISOs specifically so that you can ETTR?

Note: For anyone else answering, please take it as a given that the ideal is to get as much actual light into the camera as possible. I am interested in situations (in my experience, most situations) where something has to give - you need a fast shutter-speed to prevent motion blur, you need to close up the aperture for more DoF – in these situations where you can't just let more light in, what is the best strategy? To forego ETTR or boost the ISO to achieve ETTR?

Answer 1

It depends on the properties of the sensor in your camera.

Raising the ISO setting means you amplify the signal before reading it out, this means your signal level is higher and thus read noise is lower relative to the signal, improving the overall signal to noise ratio.

However Sony Exmor sensors (found in all NEX bodies, and many current Nikon/Pentax models but not Canon) and some newer Panasonic sensors have such low read noise that it doesn't really matter if you boost the signal before readout or not.

With such sensors there is little benefit to using the ISO setting at all when shooting RAW in manual mode as all you are doing is increasing the chance of overexposing the image somewhere.

Answer 2

I also think that it depends on the camera. Factors include the bit-depth of the A/D, the various sources of noise, and unknown details.

With all the theory requiring possibly unknown parameters even if the model is right, the only thing to do is a real test.

Assuming the histgram fits completely in the exposure so you aren't deciding which end to cut off, shifting the darks up is expected to give less quantization as there are more numbers for the zone. But, if the number of discrete electons being measured is also in the same range, multiplying it won't reveal any more shades other than noise.

Note that adding more light (e.g. longer exposure) without cutting off whites is always a clear win. Cranking the ISO is what I'm lamenting on above. Assuming you don't grt blooming or crosstalk in the brightest... so try it.