Originally posted by civiletti It makes a difference only in scenes with more than 13 stops of range, but then it makes an important difference.
I suppose it depends on how much
significant info is in the 14th stop band. When test images from the K-5 first came, GordonBGood decided to see how much shadow info would be lost when he reduced the K-5's 14-bit raw to 12-bit, dithering it sufficiently. He saw little worthwhile improvement. So I'm inclined to believe that, assuming the sensor has no banding issues i.e. is clean/uncorrelated in its noise distribution, it won't make a significant difference when boosting deep shadows, whether the system offers 13 stops or 14 stops of technical DR.
And then we have veiling glare. Consider the Canon 20D with some different types of lenses:
Imatest - Veiling Glare AKA lens flare
Even with the best lens, (primes are usually better than zooms in this regard) veiling glare only got down to 0.291% i.e. -8.4 stops down. When DxOMark measured the 20D, they only achieved a max. DR of 11 stops.
Tests and reviews for the camera Canon EOS 20D - DxOMark
But if the noise floor is uncorrelated, then veiling glare may end up being a greater inhibitor of actual DR in some shooting situations.
A paper that discusses the problem of veiling glare:
http://graphics.stanford.edu/papers/glare_removal/glare_removal.pdf
Quoting:
Based on simulations, the Canon 20D can record nearly 20 stops of dynamic range using HDR imaging if only a point light source is present. If half of the field of view is
covered by an extended source, then only 9 stops of dynamic range can be recorded by the 20D.
Here is message describing an 2008 GoogleTech video of a talk by John McCann of Polaroid: