As far as ISO ratings goes, there's been a lot of discussion about noise on LuLa lately...I suggest doing a search to read up. As far as "real ISO", Canon is finally up to 3200 with the 5D MII but Nikon's D3 is at 6400. Anything other than those are false ISO that amounts to under exposing at the next lowest real ISO and over-processing.
"Real ISO" depends on the definition of ISO. With a saturation based ISO (see
here), exposure of an 18% reflectance target with normal metering will yield a saturation of 12.5%, leaving 0.5 EV of headroom. At the lowest ISO, saturation is determined by the sensor itself and one could say that this is the true ISO of the sensor. When one sets the camera to a higher ISO, the sensor does not collect more electrons, but the gain of the camera amplifier is increased and saturation (overflow) occurs in the analog to digital converter (ADC). This is analogous to push processing of film. With the "fake" high ISOs, maximum amplifier gain has been reached, and the raw file is tagged with instructions to increase the "exposure" in post processing.
If cameras had better electronics, it would not be necessary to increase ISO with a camera setting, but one could achieve the same effect in post processing. This approach is already used with some medium format digital cameras. With current dSRLs, it is often advantageous to increase the ISO with a camera setting up to a certain limit. Beyond that limit, one can increase the ISO in post processing and gain extra headroom for the highlights. For example, with the Canon 1D MII, it makes little sense to increase the ISO beyond 1600 (see the excellent analysis by
Emil Martinec (the section entitled "S/N and Exposure Decisions"). One could question if the 1D MII really has an ISO of 3200.
The images obtained at ISO 1600 with the 1D MII might not appear to be exposed to the right and one theoretically would lose levels in the upper f/stops of the dynamic range of the image. However, as Emil explains, the real rationale of exposing to the right is to obtain a better signal to noise ratio and not make better use of the number of raw levels available in the raw data. At high ISOs the effective number of levels is determined by noise even with a 12 bit ADC. With a 14 bit ACD, the extra levels are mainly used to quantize shot noise.
Bill