Quote Originally Posted by Oren

Ok Daniel, I've just taken 2 pictures both with the same shutter speed and aperture but one with ISO 100 and the other with ISO 1600, but not surprisingly, they look very different - exposure wise, how do you expect me to compare such different pictures?
With nerdy image analysis tools, of course! Just kidding. Although the noise can be measured directly without adjusting the images, human eyeballs are better at seeing things when they are the same brightness. Just increase the brightness of the ISO 100 shot by +4 EV to match ISO 1600.

(By the way, they have the same "exposure" but different "brightness". The idea that ISO is a part of exposure is a common misconception.)




Quote Originally Posted by Jon Ruyle
Hmmm. I wonder if they could get even lower read noise by pusing that even further... maybe get rid of iso 200, 800, 3200 and leave only 100, 400, 1600.
The thing is that all the 1-stop ISO are already performed by only one single amplifier. (Just as all the tweener ISO are performed by one amplifier, making a total of two amplifiers.)

Emil has a chart of the read noises (including tweeners) in his Noise, Dynamic Range and Bit Depth in Digital SLRs essay:





Quote Originally Posted by Jon Ruyle
As an aside, I believe that read noise always decreases less than linearly.
Correct.

Quote Originally Posted by Jon Ruyle
That is, you'll always get lower noise (but lower dynamic range, too) by doing a single low iso exposure than several high iso ones (ie, a single 16 sec iso 100 exposure will have less noise than 16 1 sec iso 1600 exposures averaged out or added together).
That's correct. Read noise adds in quadrature, but it still adds.