Good thing you're on the job, Pete!  Really, if you don't have anything to
add to a thread except a spelling correction, why don't you go on out and
get a life instead?

--
www.mattclara.com

What a lovely man you are, niceparking... oops, I mean Matt Clara.
What did 'nice' think about it? - he hangs out on the HP groups too,
what a coincidence that is..

FTR, Canon DO have a reputation for the lowest noise sensors/
processing.  Did you not know that?  Do we really have to post links?

I didn't detect any smugness, and your prolific and unnecessary
insults are on display as usual.  Do you perhaps have Tourette's
syndrome, or some other medical condition that alters your brain
state?  If so, I guess we should be sorry for you, but you really need
to develop some strategies for limiting your access to social
interaction at those 'bad' times.

If not, well, it seems clear who is the "a.shole".  I mean, even Rich
gave a reasonable, if not comprehensive interpretation..

Takes one to know one, Matt.  And my answer was in direct response to
your remark "or Canon's noise is awful!"

Or don't you like anyone responding to the crap you post?

Colin D.

By examining RAW data.

There are clues about what the camera is doing, a lot like archaeology.

For the 30D, the RAW histograms have spikes (doubled values) that show
that two different values became one, periodically.  The evidence clearly
suggests that ISO 160 was originally ISO 200, and the resulting digitized
values were obtained by multiplying an original set of digitized values
by 100/126.  A second verification is the level of blackframe noise; it
is exactly 100/126 of what it is at ISO 200, at ISO 160.  No such linear
correlation exists in the 100/200/400/800/1600 base set of ISOs.  For ISO
125, the opposite clearly happens; there are gaps (unrepresented values)
in the RAW histogram that clearly show that the original digitized values
were  multiplied by 126/100, and the blackframe noise is exactly 126/100
what it is at ISO 100.  If you value your shadows, you need to know these
things.  The extreme highlights are affected, too.  All the in-between
ISOs have 1/3 stop less RAW highlights, relative to metered grey.

For the 1D* and 5D cameras, the alterations seem to be based on pushing
exclusively (not pulling and pushing like the 30D).  Unlike the 30D,
however, the extra gain is analog, but not optimized in a trend like the
base ISOs, so even though the RAW histograms look the same at all ISOs on
a given camera, spike- and gap-wise, and they all have the same headroom
relative to metered middle grey, ISOs 125 and 160 have more read noise
than ISO 200.  ISOs 160 and 320 have more than ISO 450, etc, but the
trend flattens out more at 1250 and 1600, because the relationship
between ISO and read noise becomes more linear at ISOs 800 and 1600.  
IOW, the biggest losses in quality due to extra read noise occur at the
lower "extra" ISOs.

The 160/320/640 set are the worst on the high-end cameras, and the best
on the 30D, because of the different ways they are implemented.  If I had
a 30D, and I shot JPEG exclusively, I would use the 160/320/640/1250 set
exclusively, as they give the lowest noise, and the 1/3 stop of headroom
lost relative to the metering is always discarded in JPEGs, anyway.  
Canon does not seem to recognize the top 1/3 stop of RAW data DR in the
green channel, and what is at that level or higher in the other two
channels after white-balancing.  The 160/320/etc series on the 30D pull
that other-wise wasted 1/3 stop of DR into use in the JPEG.