Let me take another crack at explaining it.



Here is how most people understand high ISO:
  • When I use autoexposure mode.
  • At low ISO, the noise is very little.
  • At high ISO, the noise is very high.
  • Therefore, high ISO causes noise (incorrect).



The
logical error is that correlation is not causation. In autoexposure
mode, high ISO reduces the exposure. It is the reduced exposure that
causes the noise to be very high. Not the high ISO. In fact, if you had
left ISO the same (low ISO) and just reduced exposure manually, the
noise would be even worse than high ISO. This is because high ISO
actually *reduces* noise. Of course, it doesn't reduce it enough to
make up for the reduced exposure, but every little bit helps.



Here is the correct way to understand high ISO:
  • When I reduce exposure, it causes more noise.
  • Sometimes when I reduce exposure, I am left with more highlight headroom than I need.
  • In those cases, I can trade away the highlight headroom to get less noise by using high ISO.
  • This
    is all thanks to the fact that the 50D sensor technology has an
    advanced feature where the actual read noise is reduced at high ISO.
    Some other cameras, such as MFDB, don't perform that way.



To prove this to yourself, just take two pictures with the same exposure (i.e. same f-number and
shutter). Set one to ISO 100 and the other to ISO 1600. Then examine
the shadows of both images and tell me which one has more noise. (Of
course there is a difference in blown highlights, which is why I
explained "ETTR then ITTR".)


Decreasing exposure causes an increase in noise. And people tend to
use high ISO at the same time as they decrease exposure, but that
doesn't mean the ISO caused the noise. If it did, then keeping the
exposure the same and increasing ISO would increase noise. It doesn't.
It decreases noise (and blows highlights).



This is one case where the wide use of autoexposure causes many folks to look at things incorrectly.
Saying "high ISO causes noise" is similar to saying "high ISO causes
thin DOF". Everyone knows that's not correct. High ISO *tends* to be
used with wide f-numbers, and wide f-numbers tend to result in thinner
DOF, but that doesn't mean the thin DOF was caused by ISO. In the same
way, noise is not caused by ISO, it is caused by underexposure.



Remember that the least amount of noise results from "ETTR then ITTR." That means to
expose to the right and only then increase ISO (if possible without
blowing important highlights). If someone did the opposite (increase ISO instead of increasing exposure), it would result in far more noise: that would be "ITTR then ETTR", which is the opposite of what I am describing.


Most people have their own home-grown and incorrect
explanations why low ISO has more noise than high ISO when exposure is
fixed. The real reason is that the sensor has physically lower read
noise at higher ISO.



Of course, ideally, we would like low ISO to have the same low read
noise that high ISO has. If it did, there would never be any
reason to use high ISO. A digital push of low ISO would give you the
exact same image as a high ISO shot, but it would have many more stops
of highlight headroom (more dynamic range).



In fact, some cameras are like that, including most MFDB. But the 50D is different. That is why
understanding how high ISO *reduces* noise is important.