I can't agree with the argument that "the processor simply manipulates the exposure to achieve the intermediate setting." Adjusting ISO doesn't change the exposure all, it's just math performed on recorded pixel values. By definition, exposure is exposing the sensor to light, which is only moderated by aperture and shutter speed. (And perhaps filters, lens design, etc.) I'll admit that theoretically I could see 1/3 stop ISO adjustment having a minuscule impact due to rounding errors, but only on very dark pixels with low values, and not enough to really matter.
If you're interested in the math behind photography, read on! (If not, skip to the TLDR
)
Think of it this way, ISO value is the adjustment of the original readout for a pixel on the sensor in terms of percentage, so ISO 100 is 100%, ISO 200 is 200%, etc. Stops are a method of doubling, so the relationship between 'stops' and light is exponential rather than linear. A full stop is simply doubling the value and only uses whole numbers. 1/3 stops are in between and require decimal adjustments, so a pixel with a value of 1 would adjust to 1.26(+1/3), 1.59 (+2/3), and 2 (full stop). Continuing on, the ISO adjustments for a pixel starting with a value of 1 between ISO 100 and ISO 12800 would be:
1, 1.26, 1.59,
2, 2.52, 3.17,
4, 5.04, 6.35,
8, 10.08, 12.70,
16, 20.16, 25.40,
32 40.32, 50.80,
64, 80.63, 101.60,
128, and so on (bold numbers are full stops, 1/3 stop decimals are approximate).
For simplicity, these are usually rounded to be easier to remember off the top of your head when writing them out.*
1, 1.25, 1.60,
2, 2.5, 3.2,
4, 5, 6.2,
8, 10, 12.5,
16, 20, 25,
32 40, 50,
64, 80, 100,
128, and so on
These decimal pixels then have to be rounded to whole numbers when they are recorded, causing a slight error. As the pixel's value increases, the impact of the rounding error becomes significantly less important. (see
signal to noise ratio) When you think about how a 12-bit image can have 4096 different values for each raw pixel, and a 14-bit image can have different 16384 values for each raw pixel, this would mean that 1/3 stop ISO adjustments only really affect the very darkest portions of the images. (At least, in any way we could see with our very non-numerical eyes!)
However, this could all be a moot point if ISO adjustment is performed within analog circuitry before being converted to bits and bytes in the analog/digital converter. If it's an analog ISO adjustment I still can't see how full or 1/3 stop adjustments would be better or worse than each other unless there are problems with the analog circuits. An analog ISO adjustment would probably be more accurate than adjusting digital values, though it would still be susceptible to noise and the end results would still be limited to bits and bytes.
TLDR: Yes, theoretically I can see that 1/3 stop ISO adjustments can cause rounding errors in individual pixels, but would only really affect very dark pixels with low values, and the error introduced will likely be so small that we would never be able to see any negative effects on our images.
*You might have noticed this is similar to the progression of shutter speeds, though again they're rounded to make them easier to remember...
1, 1/2, 1/4, 1/5, 1/6,
1/8, 1/10, 1/13,
1/15, 1/20, 1/25,
1/30, 1/40, 1/50,
1/60, 1/80, 1/100,
1/125, 1/160, 1/200,
1/250, 1/320, 1/400,
1/500, 1/640, 1/800,
1/1000
(Remember, 'Stops' are just operations of doubling or halving values!)
---------- Post added 05-24-16 at 03:44 PM ----------
Originally posted by robthebloke In my very unscientific experiments with my K-1, adjusting the ISO in camera produces less noise than compensating exposure later in lightroom (but it is a very close run thing).
Same here on my K-3. This would lead me to believe that in camera ISO adjustment might be done in analog circuitry before being converted to digital.
Analog can be more accurate than digital since it can allow for intermediate values while digital works with whole numbers (in binary).
Here is an example converting four hypothetical pixel values from ISO 100 to ISO 1600, starting in analog and ending in digital.
1, 2.5, 3.2, 4.6 in analog adjusted to ISO 1600 -> 16, 40, 51.2, 73.6 -> then converted to digital: 16, 40, 51, 74
1, 2.5, 3.2, 4.6 in analog converted to digital -> 1, 3, 3, 5 then adjusted to ISO 1600 -> 16, 48, 48, 80
You can see that ISO adjusting in analog can be more precise, but adjusting in digital only give end values in increments of 16, so there is less differentiation between pixels.