In the spirit of "seeing for myself", I took a series of images using my 5D Mark II camera at different ISO settings.

Here is the full test image. This is a 5616x3744 pixel image downsized to 800x533. I put my 5Dii on a tripod with the 100mm f/2.8 macro lens, set to f/16. I shot tethered and pulled the raw files directly into Lightroom 5. I did the conversion to JPEG using default settings, no sharpening or adjustments.

The following images are 640 by 480 pixel crops pulled from the 5616 by 3744 pixel images. So they show the noise at 1:1 with no sharpening or contrast adjustment of any kind. It should be obvious what part of the above photo these represent. Someday when I learn some new tricks, I will draw a box on the above photo to make things even more clear.

The subjects are a seed pot, approximately 3 inches in diameter from the Acoma Pueblo in New Mexico and a size 16s pocketwatch.

Conclusions

I now intend to use ISO 400 on a routine basis, and I won't feel much hesitation about using ISO 800.

ISO 25600 looks pretty bad on these crops, but for a full screen image (or something like an 8x10 print), it is quite usable, and not really any worse than we used to get with 8x10 prints from ASA 400 film. See the pair of images at the end of this page which shows an image of typical "web illustration size" taken at ISO 25600.

I would be even more likely to use ISO 6400 in a pinch, which is significantly better than ISO 25600.

I skipped ISO 50, for reasons I won't publicly admit to. However given the noise performance at ISO 100 and 200, I don't feel that any important information was lost.

There are some surprising fine points as well. Apparently ISO 50 is derived in the camera by running the analog circuitry at gain settings for ISO 100 and then dividing by 2 in software. Similarly, ISO levels above 3200 are derived digitally after the analog circuitry, or something like that. If you examine images with the right software ("rawanalyze") you will see missing codes that confirm these statements. In essence what is done at ISO 6400 is to underexpose by a stop and then multiply the codes by 2 digitally - the fact that all codes are even numbers reveals this. You can shoot raw at ISO 3200 and then do this yourself.

ISO 100 and 200 follow:

ISO 400 and 800 follow:

ISO 1600 and 3200 follow:

ISO 6400 and 12800 follow:

ISO 25600 follows: