I downloaded the raw files from the akihabaranews site and made some preliminary analysis. Preliminary not because the camera is a preproduction copy (it is naive to believe, that the sensor will be changed of a camera, which hits the shops in two-three weeks). Preliminary, because the shots are only limitedly suitable for such objective analysis.
I measured the per pixel noise in terms of standard deviation on selected spots in a selected raw channel. It is important to understand, that it plays no role, which channel is picked, only the pixel values are of meaning in this aspect.
The first observations are:
- the top real ISO is 3200; compare this to 5d: 1600, Nikon D3: 6400, 1DMkIII; 6400, 1DsMkIII: 3200, 40D: 1600, 50D: 3200;
- the 1/3 EV ISOs are fake, like with the 40D and 50D, unlike the Nikon D3, the Canon 1DMkIII (and probably the 1DsMkIII). Though this is irrelevant for raw shpooters;
- the per pixel noise is practically equal to the 5D at ISO 400 and
slightly lower than the 5D at ISO 1600;
- compared to the Nikon D3: they are so close, that the results of measurements are somewhat ambiguous.
Some scepsis about the real usefulness of higher ISO gain is justified. Sensor manufacturers can increase the gain until it "catches" a single electron, but does the gain contribute to the image content with details (better distinction between details with almost identical intensity), or will only the noise be increased?
My first observations based on these 5D2 raw files are that
1. the step from ISO 800 to 1600 yields a big real gain,
2. the step from ISO 1600 to 3200 is almost worthless.
(As posted already earlier, ISO 6400 does not yield any real gain).
To 1: I don't have any ISO 800 shot, but I do one with ISO 640, which is the same as ISO 800; however, it has been shot not 1 EV but 1 1/3 EV higher than the ISO 1600 shot (of course), thus the DRP adjustment has to be 1.33.
Example:
5D2 ISO 640 DRP 8.31 NR 26A spot with the same amount of captured light appears in the dynamic range position 1.33 EV higher in the ISO 1600 shot.
Example: ISO 1600, DRP 7.01, NR 20; *the ISO gain reduced the noise a lot*:
5D2 ISO 1600_DRP 7.01NR 20To 2:
a) Both shots were made @ f/6.3, the ISO 3200 with 1/20s, the 1600 with 1/10s, The histograms show, that these shots are directly comparable:
5D2 ISO 1600 Histogram5D2 ISO 3200 Histogramb) The noise-to-signal ratio (NR) in the ISO 1600 shot at the dynamic range position (DRP) 7 is about 20%; I found this in several selections, one example is posted just above.
c) Pixels with the same amount of captured light appear in the ISO 3200 shot 1 EV intensiver ("brighter"), i.e. at DRP 6. Several selections show, that the NR is 19%, i.e. the one stop higher ISO managed to reduce the noise to signal ratio from that fixed amount of light only 5%; *this is meager*. One example for such a selection (keep in eye: the color is irrelevant; the subject is not the light coming through the lens but the light captured by the sensel);
5D2 ISO 3200 DRP 6 NR 19d) Only for comparison: the NR in a shot with ISO 3200 at DRP 7 (i.e. half the light compared to the selection with DRP 6) is 32%:
5D2 ISO 3200 DRP 7 NR 32The analysis has been done by
Rawnalyze