The Signal to Noise (S/N) ratio in Peter's charts tells you how much better the maximum signal is relative to the noise floor, or in other words (with a theoretically constant signal) how noise deteriorates the ratio. A higher S/N ratio is better so, the 30D's noise floor being the lowest at ISO 200, the S/N ratio is highest (best quality in the shadow areas).
You meant 160, didn't you? That has the highest value on the 30D chart.
These aren't really S/N ratios, though. S/N depends on the signal and noise at various signal levels; The figures in these charts are the ratios of the maximum signal, to the noise of
no signal at all. They should be called DR, more properly, and even then, they are slightly different from the ratios of max signal to the signal where the S/N is equal to 1:1, because there is a small amount of shot noise from that perspective, relative to read noise. The ratios on these charts are not affected by shot noise, they are like classic audio S/N where there is no such thing as shot noise.
With the native sensitivity of current Bayer CFA sensor arrays being approx. equal to ISO 100,
I wonder if there is some industry convention to have ISO 100 if at all possible, and have the extended main ISOs to be full stops faster? It seems that cameras are compromised to have these round numbers - ISO 50 is compromised on the 5D and 1D* cameras; ISO 100 on the 20D/30D, etc. It would seem much better, IMO, to have the lowest ISO at whatever sensitivity is necessary to have, say, 3.5 stops of RAW green highlights in the linear range of the sensor, below saturation, and then have other ISOs built upon this base, all with the same headroom. SO, the 1D* and 5D would have ISO 80, 160, 320, 640, 1280, 2560, (5120,) etc. The 20D and 30D could start at ISO 125. It seems like so much is twisted to have smooth-number ISOs, and much more than semantics are damaged. The improvement in shot noise by having ISOs 50 and 100 in these cases is minor, and only happens when one trusts the camera's meter for a centered histogram; all the benefits are possible by just exposing to the right at hypothetical main ISOs 80 and 125, with higher potential DR for the base ISO.
it suggests that the 30D uses a combination of 'pushing' and 'pulling' for intermediate ISO settings. This is different from other models which only 'push' intermediae ISOs.
The way in which the 5D and 1D cameras push intermediate ISOs is actually the way *all* ISOs are pushed on most CCD sensors, AFAIK. Looking at noise levels in Nikon files, for example, read noise is almost directly proportional to ISO, and ISO 1600 typically has about 15x the read noise of ISO 100, suggesting that the only difference in absolute read noise (IOW, in electrons) amongst ISOs (other than quantization), is the ADC noise.
You won't see jagged charts with these cameras, because it is all scaled, and there are no optimization for a higher ISO at any point on the chart. You won't see any ratios higher than about 650, for example, at ISO 100 on a D2X, and ISO 1600 is only going to be about 40, and a smooth curve in-between. With cameras that sport an almost linear relationship between ISO and read noise, there is not as much benefit in exposing to the right if one shoots RAW, as shooting to the left at a lower ISO can give similar results.
So, there are actually many levels of ISO "reality". What gets called fake in one context may be what is normal in another. They are all real, as far as exposure index is concerned (other than some small metering differences between cameras). The 5D, 1D, and most CCD cameras achieve some or all of their ISOs by some secondary amplification after the inital read and before the digitization, and this allows the headroom to be consistent in all, and the DR graphs to be smooth in the CCD cameras.
The 30D way, in general, is the most annoying of all, as the headroom varies in neighboring ISOs, as does read noise relative to metering, and stacking images from the extra ISOs, while reducing noise, will also uncover the posterization effects hidden in the gapped and spiked histograms. The 30D does have one benefit, though. The 160/320/640/1250 group allow a virtual extra 1/3 stop positive EC, and they automaically use 1/3 stop more DR than the main ISOs for JPEGs. My 30D is almost always set to this group of ISOs (especially at the lower end of the ISO scale), and I'd only use 200, 400, etc if I were stacking or binning RAW images and didn't want to suffer any posterization in the deepest shadows.
