What does a fractional colour bit depth mean?

Question

Looking at camera comparison sites today, I encountered statements like "color depth: 23.5 bits".

(See here for example, which gives the 5D Mk III a 24-bit colour depth, and the D7000 a 23.5 bit colour depth).

My usual understanding is that in a 24-bit colour representation, there are 8 bits per channel (i.e. 8 bits for each of the red, green and blue channels). Given that a bit is indivisible, how do I interpret a fractional bit depth? It this just somebody taking log2(max-color-value)** and forgetting to round/truncate, or is it meaningful in terms of hardware or image encoding (or something else)?

** e.g. if the actual sensor colour space is smaller than a 24-bit RGB colour space, and then the effective depth is log2 of that. But the pixels are 14-bit in most high-end cameras, so this doesn't make sense either...

Answer 1

Snapsort is using DxOMark for its sensor data, and DxOMark explain here what each of their scores mean. Specifically for Color Depth, they say:

Maximum color sensitivity reports, in bits, the number of colors that the sensor is able to distinguish.

i.e. DxOMark claim that the 5D Mark III can distinguish 2^24 = 16.8 million colours, while the D7000 can distinguish 2^23.5 = 11.9 million colours. Don't worry too much about the absolute numbers, all it means is that the 5D Mark III can distinguish about √2 times as many colours as the D7000, at least by DxOMark's metrics (which they don't make public).