An interview with Canon:

http://www.imaging-resource.com/news...e-6D-and-whats

The spin:
"We very much feel that that resolution, 20 to 22 megapixels on a full
frame sensor is the real sweet spot. It's the maximum that we can get to
without starting to see a conflict between noise performance and
resolution. And obviously, what we want to try to do is give you the
best possible balance. And over the last few years, what you've seen is
that other manufacturers have been sort of creeping up with their
resolution. But we've reached a sort of sweet spot since 5 years ago
when we launched the 1DS Mark III. And we think that, at the moment,
that is the better place to be in terms of performance, allowing people
to have the high ISO capabilities--high as you're able to get with the
current Canon range as well as the high resolution and image quality."

The probable truth:
Canon doesn't have the technology to stitch higher resolution 35mm
format CMOS sensors with the precision required to include on-chip a/d
conversion. They've more-or-less plateaued with read noise, so if they
increase pixel count, they're stuffed.
As he says - "It's the maximum that *we* can get".


So much promise:
http://www.robgalbraith.com/public_f...hite_Paper.pdf

Me wrote:

The spin:
"We very much feel that that resolution, 20 to 22 megapixels on a full
frame sensor is the real sweet spot. It's the maximum that we can get to
without starting to see a conflict between noise performance and
resolution. And obviously, what we want to try to do is give you the
best possible balance. And over the last few years, what you've seen is
that other manufacturers have been sort of creeping up with their
resolution. But we've reached a sort of sweet spot since 5 years ago
when we launched the 1DS Mark III. And we think that, at the moment,
that is the better place to be in terms of performance, allowing people
to have the high ISO capabilities--high as you're able to get with the
current Canon range as well as the high resolution and image quality."

Canon's not playing the megapixel race. This is a good thing.

The probable truth:
Canon doesn't have the technology to stitch higher resolution 35mm
format CMOS sensors with the precision required to include on-chip a/d
conversion. They've more-or-less plateaued with read noise, so if they
increase pixel count, they're stuffed.
As he says - "It's the maximum that *we* can get".

Canon has been consistently lower in per pixel read noise than
Nikon.[1] Nikon's even cheating in that area (clamping values
to zero instead of using an offset). Astrophoto people are less
than happy that very faint signals get wiped out between sensor
and RAW file and therefore won't add up to a faint star when they
stack their exposures.

So ... your assumption that the problem is read noise is busted.


-Wolfgang

[1] http://clarkvision.com/articles/digi...mance.summary/
The 5D2 (i.e. the old one) has a read noise of 2.5 e¯ at
higher ISOs.

http://home.comcast.net/~NikonD70/Ge...Read_Noise.htm
D800 has a read noise of 4.1-4.2 e¯ at higher ISOs
(Read noise is worse at lower ISOs)

In addition, the D800 switches to "digital" gain (i.e.
multiplication) after ISO 1000.

On Mon, 24 Sep 2012 20:28:02 +0200, Wolfgang Weisselberg
wrote:

Me wrote:

The spin:
"We very much feel that that resolution, 20 to 22 megapixels on a full
frame sensor is the real sweet spot. It's the maximum that we can get to
without starting to see a conflict between noise performance and
resolution. And obviously, what we want to try to do is give you the
best possible balance. And over the last few years, what you've seen is
that other manufacturers have been sort of creeping up with their
resolution. But we've reached a sort of sweet spot since 5 years ago
when we launched the 1DS Mark III. And we think that, at the moment,
that is the better place to be in terms of performance, allowing people
to have the high ISO capabilities--high as you're able to get with the
current Canon range as well as the high resolution and image quality."

Canon's not playing the megapixel race. This is a good thing.

More correctly, Canon *can't* play the megapixel game. Sony and others
have shown that you can resolution without adding too much noise.
Canon can't with their current technology so it's true that *Canon*
have plateaued, others have not.


The probable truth:
Canon doesn't have the technology to stitch higher resolution 35mm
format CMOS sensors with the precision required to include on-chip a/d
conversion. They've more-or-less plateaued with read noise, so if they
increase pixel count, they're stuffed.
As he says - "It's the maximum that *we* can get".

Canon has been consistently lower in per pixel read noise than
Nikon.[1] Nikon's even cheating in that area (clamping values
to zero instead of using an offset). Astrophoto people are less
than happy that very faint signals get wiped out between sensor
and RAW file and therefore won't add up to a faint star when they
stack their exposures.

So ... your assumption that the problem is read noise is busted.

Whatever. I take pictures, not science tests. It's clear that Canon is
behind in sensor tech right now. That may change, but right now Nikon
is producing cameras with better sensors.

Bowser wrote:
On Mon, 24 Sep 2012 20:28:02 +0200, Wolfgang Weisselberg
Me wrote:

The spin:
"We very much feel that that resolution, 20 to 22 megapixels on a full
frame sensor is the real sweet spot. It's the maximum that we can get to
without starting to see a conflict between noise performance and
resolution. And obviously, what we want to try to do is give you the
best possible balance. And over the last few years, what you've seen is
that other manufacturers have been sort of creeping up with their
resolution. But we've reached a sort of sweet spot since 5 years ago
when we launched the 1DS Mark III. And we think that, at the moment,
that is the better place to be in terms of performance, allowing people
to have the high ISO capabilities--high as you're able to get with the
current Canon range as well as the high resolution and image quality."

Canon's not playing the megapixel race. This is a good thing.

More correctly, Canon *can't* play the megapixel game.

Do you have proof? And what's this:
http://www.canon.com/news/2010/aug24e.html
then?

Sony and others
have shown that you can resolution without adding too much noise.

Have they?

Let's look at
http://www.dpreview.com/reviews/sonyslta77/14
Sony's SLT-A77, a 24 MPix crop versus the (older) 7D (18 MPix)
and the Pentax K-5 (16 MPix).

Look at the RAW noise comparison, the graph.
Oh, the A77 is quite a bit higher, but underreports the ISO by
1/3rd stop. (And the K-5 is really low, but the reported ISO
values are 1/3rd stop too low.)

Calculate that you could (theoretically) bin the pixels to 16 MPix:
24/16 pixels give a factor of SQRT(1.5) == 1.22 less noise and 18:16 pixel give
a factor of SQRT(1.125) = 1.06 less noise.

Let's check at ISO 100, 400, 1600 and 6400 (each adjusted for
it's ISO reporting shift) for colour and black noise ... and
correct for the smaller pixels:

Chroma noise:
ISO 100 400 1600 6400
A77 2,9 5,3 10,6 14,7
7D 2,8 5,2 9,9 11,3
K-5 2,0 4,0 7,5 9,0

Black noise:
ISO 100 400 1600 6400
A77 2,4 3,7 8,2 14,7
7D 1,9 2,8 6,6 11,3
K-5 2,0 2,5 5,5 9,0

The values should be correct to ± 0.5.

There is noticeably more noise *even in images of the same print
or web size* at higher ISO. When pixelpeeping, it's even worse!

So any advantage of the A77's slightly higher linear resolution
comes only to pass when you display or print in a size where
that extra resolution means no or less interpolation of pixels.
But you pay for it with noisier images ... every time.

If you want low noise (and a tad less resolution), the K-5's
the idea.

If you want a resolution to really matter, start printing HUGE.


Canon can't with their current technology so it's true that *Canon*
have plateaued, others have not.

Can you prove that that's so? Have any leaked strategy papers
from Canon? Slept with someone high in the development unit and
had interesting pillow talk?


The probable truth:
Canon doesn't have the technology to stitch higher resolution 35mm
format CMOS sensors with the precision required to include on-chip a/d
conversion. They've more-or-less plateaued with read noise, so if they
increase pixel count, they're stuffed.
As he says - "It's the maximum that *we* can get".

Canon has been consistently lower in per pixel read noise than
Nikon.[1] Nikon's even cheating in that area (clamping values
to zero instead of using an offset). Astrophoto people are less
than happy that very faint signals get wiped out between sensor
and RAW file and therefore won't add up to a faint star when they
stack their exposures.

So ... your assumption that the problem is read noise is busted.

Whatever.

Ah, got it. Everything's fine if one can put down Canon,
truth or reality doesn't matter.

I take pictures, not science tests.

In which case you don't need the extra resolution and will be
eminently fine with 20 MPix.

It's clear that Canon is
behind in sensor tech right now. That may change, but right now Nikon
is producing cameras with better sensors.

And if you're taking pictures and not science tests, the sensor
quality differences between Nikon and Canon don't matter to you.

However, the fact that Canon has managed to nail PDAF as accurate
as CDAF (and has lenses and bodies on the market that can pull
that stunt) may be very important to your pictures.


-Wolfgang

On 25/09/2012 6:28 a.m., Wolfgang Weisselberg wrote:
Me wrote:

The spin:
"We very much feel that that resolution, 20 to 22 megapixels on a full
frame sensor is the real sweet spot. It's the maximum that we can get to
without starting to see a conflict between noise performance and
resolution. And obviously, what we want to try to do is give you the
best possible balance. And over the last few years, what you've seen is
that other manufacturers have been sort of creeping up with their
resolution. But we've reached a sort of sweet spot since 5 years ago
when we launched the 1DS Mark III. And we think that, at the moment,
that is the better place to be in terms of performance, allowing people
to have the high ISO capabilities--high as you're able to get with the
current Canon range as well as the high resolution and image quality."

Canon's not playing the megapixel race. This is a good thing.

The probable truth:
Canon doesn't have the technology to stitch higher resolution 35mm
format CMOS sensors with the precision required to include on-chip a/d
conversion. They've more-or-less plateaued with read noise, so if they
increase pixel count, they're stuffed.
As he says - "It's the maximum that *we* can get".

Canon has been consistently lower in per pixel read noise than
Nikon.[1] Nikon's even cheating in that area (clamping values
to zero instead of using an offset). Astrophoto people are less
than happy that very faint signals get wiped out between sensor
and RAW file and therefore won't add up to a faint star when they
stack their exposures.

So ... your assumption that the problem is read noise is busted.


Hmmm....
So, /minimum/ read noise of the 5DIII is actually even better 2.4 e-
Compared to D800 2.6 e-, this should be even better still?
Sorry, 5DIII read noise measured at 2.4 e- is at (Set) ISO 12,800 (=
real ISO 10,084 - Canon don't tell the gospel truth about ISO either -
all darned camera makers are prone to exaggerate it seems). Compared to
shot noise at this sensitivity setting this read noise is like farting
against thunder.
Base ISO 100 (set on camera) read noise is:
5D III 33.1 e-
D800 2.7 e-
It's a very very big difference.
"Read noise is worse at lower ISOs" is very true for Canon, and very
much less so for Nikon/Sony Exmor.
You posted a link above to Bill Claff's site. His "photographic dynamic
range" is explained he
http://home.comcast.net/~nikond70/Ge...amic_Range.htm
I think that's a pretty good method from which he presents a chart
application, for example comparing the Nikon D800 and Canon 5D III he
http://i48.tinypic.com/fa55yd.png
Just for interest sake, I included a plot for "Ideal FX" - which is the
perfect 36x24mm sensor, with assumed 100% quantum efficiency (QE) and no
read noise at all.
The 5DIII has improved in (*1) QE over the 5DII, from 33% to 49%, Nikon
D800 is 56%. Anyway, I hope this is convincing evidence that read noise
(at ISO settings where it's relevant) is /much/ higher with the Canon 5D
III (Canon's best FX sensor to date) than the D800, and if this is
quantified using Bill Claff's "Photographic" Dynamic Range measure (ie
s/n ratio as would be seen in a real print at any given standard size)
then the D800 ****es all over the 5DIII (sorry to be so brutral).
Then something else appears...
The 5DIII plot on the chart "takes a turn" at about ISO 25600, it starts
"improving". Oops - by ISO 102400 it's improved so much that it's got
better than 100% QE, and zero read noise. Now this is quite something,
on a par with finding faster than light neutrinos, Canon have broken the
laws of physics. Hmmmm... perhaps it's time for a shave with Occam's
razor... - yes, once you cut the bull**** away, Canon are now applying
noise reduction to raw data /before/ writing raw files. They didn't
tell anybody that they were doing this, and apart from the curious
beginning of a non-linear curve at ISO 25600, there's plenty of evidence
that Canon are "cooking" their high ISO raw files - but actually no
evidence that they aren't also cooking raws at low ISO.


So, compared to the 5DII, what do we see with the "improved" 5DIII. A
small increase in QE, from 33-49% - nice and probably NOTHING to do with
the sensor itself, but improvements to the filter stack. High ISO seems
a bit improved, BUT they are cooking raw files. There's practically NO
difference in read noise from 5DII to 5DIII,

If this doesn't tell you that Canon have hit the wall quite a few years
ago with their sensor fabrication technology, then I have a bridge to
sell you.

Still capable of taking nice photos - base ISO dynamic range of the
5DIII is about the same as a 2007 model Dx sensor Nikon D90 which are
pretty good little cameras - you can buy a used one for about $350.


*1
How much QE is effectively lost by one of the other little quirks of
digital sensors when used with large apertures, as uncovered by DXO he
http://www.dxomark.com/index.php/Pub...s/F-stop-blues

Me wrote:
On 25/09/2012 6:28 a.m., Wolfgang Weisselberg wrote:
Me wrote:

Canon has been consistently lower in per pixel read noise than
Nikon.[1] Nikon's even cheating in that area (clamping values
to zero instead of using an offset). Astrophoto people are less
than happy that very faint signals get wiped out between sensor
and RAW file and therefore won't add up to a faint star when they
stack their exposures.

So ... your assumption that the problem is read noise is busted.

Hmmm....
So, /minimum/ read noise of the 5DIII is actually even better 2.4 e-

We can assume that Canon managed to at least hold the read noise
of the 5DII in the 5DIII.

Compared to D800 2.6 e-, this should be even better still?

Why? It's not photon noise. It's amplifier noise, A/D
converter noise and similar. It has nothing to do with the
size of the pixels.

Sorry, 5DIII read noise measured at 2.4 e- is at (Set) ISO 12,800 (=
real ISO 10,084 - Canon don't tell the gospel truth about ISO either -
all darned camera makers are prone to exaggerate it seems).

I call bull.
1. http://www.dpreview.com/reviews/cano...5d-mark-iii/18
(And they don't find that with all cameras.)
2. ISO 10,084! Not 10,083 and not 10,085 and certainly not
10,000.
The difference between 10,000 and 10,084 is that 10,000 is
0,82% lower. And 10,083 is less than 0,01% lower.

How are you to properly measure that to this exactness?
And --- how comes they report 1/3rd stop difference when
dpreview finds it's correct within 1/6th stop?

Compared to
shot noise at this sensitivity setting this read noise is like farting
against thunder.

So ... reading noise no problem? Fine! I'll come back to that!

Base ISO 100 (set on camera) read noise is:
5D III 33.1 e-
D800 2.7 e-
It's a very very big difference.

2.7 e¯ at ISO 100 for the D800? Souce, please!
What I see is
http://home.comcast.net/~NikonD70/Ge...Read_Noise.htm
4.7 e¯ already at ISO 400, and --- as always --- reading
noise in electrons increase as the ISO goes down and more
electrons are counted. Sometimes faster, sometimes slower.

And 33.1 e¯ for the 5D3? Source please!
http://www.clarkvision.com/articles/...mance.summary/
reports 30.1 and 23.5 for the 5D and 5D2 respectively at ISO
100.

"Read noise is worse at lower ISOs" is very true for Canon, and very
much less so for Nikon/Sony Exmor.

Yet how much does it matter? You're already saying that shot
noise trumps read noise at ISO 12,800. Now, ISO 100 is 7
stops slower, meaning 2^7 = 128 electrons where there was
only one before.

2.5 e¯ * 128 = 320 e¯ (5D Mark 2)

It also means SQRT(128) = 11.3... times more photon noise
(which is what causes shot noise.)

2.5e¯ * SQRT(128) = 28.28e¯ (5D Mark 2)

Compare that to the 23.5 electrons read noise of the 5D2(!)
at ISO 100. The same shot noise that --- according to you ---
trumps read noise at ISO 12,800 trumps it even more at ISO 100.

I don't see why Canon should be unable to use smaller pixels
here.

You posted a link above to Bill Claff's site. His "photographic dynamic
range" is explained he
http://home.comcast.net/~nikond70/Ge...amic_Range.htm
I think that's a pretty good method from which he presents a chart
application, for example comparing the Nikon D800 and Canon 5D III he
http://i48.tinypic.com/fa55yd.png

Why not use
http://home.comcast.net/~nikond70/Ch...,D800,Ideal_FX

Just for interest sake, I included a plot for "Ideal FX" - which is the
perfect 36x24mm sensor, with assumed 100% quantum efficiency (QE) and no
read noise at all.
The 5DIII has improved in (*1) QE over the 5DII, from 33% to 49%, Nikon
D800 is 56%.

And the QE of the 1D Mark II sensor (vintage 2004!) is already 38%.
http://www.clarkvision.com/articles/....qe/index.html

You're telling me a 2008 sensor from Canon has less QE than a
2004 sensor from the same company?


So: how do you arrive at a QE from the plot, what do you actually
measure and call QE, and how do you get to these numbers?

Anyway, I hope this is convincing evidence that read noise
(at ISO settings where it's relevant)

Nope, the lower the ISO setting, the higher the photon noise
in electrons. Hence: not relevant.

Reading noise in electrons is *much* more relevant in high ISO: the
electron count multiplication factor is doubled at every ISO stop.
Hence also the reading noise in DN (what Claff calls ADU, I think).

is /much/ higher with the Canon 5D
III (Canon's best FX sensor to date) than the D800,

Nope.

You bandied some numbers. I don't believe in merely your say-so.

and if this is
quantified using Bill Claff's "Photographic" Dynamic Range measure (ie
s/n ratio as would be seen in a real print at any given standard size)

Ah --- nope. Not the s/n ratio. At best the (in his eyes)
usable Dynamic Range. And not at any given standard size,
but at ONE print size-viewing distance equivalent. Without
indication what that may be.

And always assuming you didn't buy the D800 for the higher pixel
count, but to only print the same old sizes you printed with 8
or 15 MPix before. Once you want to actually use the higher
pixel count.

then the D800 ****es all over the 5DIII (sorry to be so brutral).

Almost any camera ****es all over any other camera if you
measure one parameter only and choose wisely.


Then something else appears...
The 5DIII plot on the chart "takes a turn" at about ISO 25600, it starts
"improving". Oops - by ISO 102400 it's improved so much that it's got
better than 100% QE, and zero read noise. Now this is quite something,
on a par with finding faster than light neutrinos, Canon have broken the
laws of physics.

Or maybe ... the way they measure the data is broken.
That's the most likely explanation.

Hmmmm... perhaps it's time for a shave with Occam's
razor... - yes, once you cut the bull**** away, Canon are now applying
noise reduction to raw data /before/ writing raw files.

You mean like Nikon does for a long time no, and not only for
extended range ISO values? Google for Nikon "mode 3" and
learn ...

And if Canon were applying noise reduction, that would be
visible in the image, wouldn't it? So ... show pix!

They didn't tell anybody that they were doing this,

.... just like Nikon ...

and apart from the curious
beginning of a non-linear curve at ISO 25600, there's plenty of evidence
that Canon are "cooking" their high ISO raw files - but actually no
evidence that they aren't also cooking raws at low ISO.

We KNOW Nikon's been cooking RAW ... at low ISO. So what's
your screaming about? It's only OK when Nikon does it?

Here's another Nikon cook:
| This is unfortunate, but it is my experience that all
| Nikon cameras do some digital scaling after the ADC.
http://home.comcast.net/~nikond70/Ge...rimer/Gain.htm


So, compared to the 5DII, what do we see with the "improved" 5DIII. A
small increase in QE, from 33-49% - nice and probably NOTHING to do with
the sensor itself, but improvements to the filter stack.

"small increase". And what do you call the increase of
linear resolution between the 5D3 and the D800 ... positively
tiny? (After all, it's only about half the small increase in
QE ...)

High ISO seems
a bit improved, BUT they are cooking raw files.

Nikon does too.

There's practically NO
difference in read noise from 5DII to 5DIII,

Source?

If this doesn't tell you that Canon have hit the wall quite a few years
ago with their sensor fabrication technology, then I have a bridge to
sell you.

If the QE increase alone doesn't tell you Canon could easily go
to 31 Mpix and get the same quality as the 5D2, I have a asteroid
to sell you, with lots of silicon --- which they really need for
computer and smartphone chips!

Still capable of taking nice photos - base ISO dynamic range of the
5DIII is about the same as a 2007 model Dx sensor Nikon D90 which are
pretty good little cameras - you can buy a used one for about $350.

Yep, if ISO dynamic range at ISO 200 is the only thing that
matters to you. As I said, choose a single stat wisely, **** on
other camera.

So why does the D800 have 36 MPix, when a 12 MPix D90 will do?

*1
How much QE is effectively lost by one of the other little quirks of
digital sensors when used with large apertures, as uncovered by DXO he
http://www.dxomark.com/index.php/Pub...s/F-stop-blues

Tell me how they measured that, when the camera does hide it.

-Wolfgang

On 28/09/2012 5:54 a.m., Wolfgang Weisselberg wrote:
Me wrote:
On 25/09/2012 6:28 a.m., Wolfgang Weisselberg wrote:
Me wrote:

Canon has been consistently lower in per pixel read noise than
Nikon.[1] Nikon's even cheating in that area (clamping values
to zero instead of using an offset). Astrophoto people are less
than happy that very faint signals get wiped out between sensor
and RAW file and therefore won't add up to a faint star when they
stack their exposures.

So ... your assumption that the problem is read noise is busted.

Hmmm....
So, /minimum/ read noise of the 5DIII is actually even better 2.4 e-

We can assume that Canon managed to at least hold the read noise
of the 5DII in the 5DIII.

Assumptions can lead to wrong conclusions. There's practically no
difference, but seeing as you mentioned it base ISO read noise is
actually very slightly worse for the 5DIII (compared to the 5DII).
Canon have "hit the wall" WRT this.

Compared to D800 2.6 e-, this should be even better still?

Why? It's not photon noise. It's amplifier noise, A/D
converter noise and similar. It has nothing to do with the
size of the pixels.

Sorry, 5DIII read noise measured at 2.4 e- is at (Set) ISO 12,800 (=
real ISO 10,084 - Canon don't tell the gospel truth about ISO either -
all darned camera makers are prone to exaggerate it seems).

I call bull.
1. http://www.dpreview.com/reviews/cano...5d-mark-iii/18
(And they don't find that with all cameras.)
2. ISO 10,084! Not 10,083 and not 10,085 and certainly not
10,000.
The difference between 10,000 and 10,084 is that 10,000 is
0,82% lower. And 10,083 is less than 0,01% lower.

How are you to properly measure that to this exactness?
And --- how comes they report 1/3rd stop difference when
dpreview finds it's correct within 1/6th stop?

http://www.sensorgen.info/CanonEOS_5D_MkIII.html
DPReview is hardly a reliable source.

Compared to
shot noise at this sensitivity setting this read noise is like farting
against thunder.

So ... reading noise no problem? Fine! I'll come back to that!

Base ISO 100 (set on camera) read noise is:
5D III 33.1 e-
D800 2.7 e-
It's a very very big difference.

2.7 e¯ at ISO 100 for the D800? Souce, please!
What I see is
http://home.comcast.net/~NikonD70/Ge...Read_Noise.htm
4.7 e¯ already at ISO 400, and --- as always --- reading
noise in electrons increase as the ISO goes down and more
electrons are counted. Sometimes faster, sometimes slower.

And 33.1 e¯ for the 5D3? Source please!
http://www.clarkvision.com/articles/...mance.summary/
reports 30.1 and 23.5 for the 5D and 5D2 respectively at ISO
100.
http://www.sensorgen.info/
Browse data to your heart's content. The figures won't tie up exactly
with Roger N Clark's, but they're close enough to see that there hasn't
been a trend of improvement at sensor level from Canon for years. (Apart
from QE, which as I said - and it was mainly a guess - probably relates
to improvements in the filter stack)

"Read noise is worse at lower ISOs" is very true for Canon, and very
much less so for Nikon/Sony Exmor.

Yet how much does it matter? You're already saying that shot
noise trumps read noise at ISO 12,800. Now, ISO 100 is 7
stops slower, meaning 2^7 = 128 electrons where there was
only one before.

2.5 e¯ * 128 = 320 e¯ (5D Mark 2)

It also means SQRT(128) = 11.3... times more photon noise
(which is what causes shot noise.)

2.5e¯ * SQRT(128) = 28.28e¯ (5D Mark 2)

Compare that to the 23.5 electrons read noise of the 5D2(!)
at ISO 100. The same shot noise that --- according to you ---
trumps read noise at ISO 12,800 trumps it even more at ISO 100.

I don't see why Canon should be unable to use smaller pixels
here.

You posted a link above to Bill Claff's site. His "photographic dynamic
range" is explained he
http://home.comcast.net/~nikond70/Ge...amic_Range.htm
I think that's a pretty good method from which he presents a chart
application, for example comparing the Nikon D800 and Canon 5D III he
http://i48.tinypic.com/fa55yd.png

Why not use
http://home.comcast.net/~nikond70/Ch...,D800,Ideal_FX
I would have done, but the link doesn't work correctly...

Just for interest sake, I included a plot for "Ideal FX" - which is the
perfect 36x24mm sensor, with assumed 100% quantum efficiency (QE) and no
read noise at all.
The 5DIII has improved in (*1) QE over the 5DII, from 33% to 49%, Nikon
D800 is 56%.

And the QE of the 1D Mark II sensor (vintage 2004!) is already 38%.
http://www.clarkvision.com/articles/....qe/index.html
Sensorgen has it at 30%.

You're telling me a 2008 sensor from Canon has less QE than a
2004 sensor from the same company?

No - it's 10% better (33%) using the same measurement methodology, but
seeing as you point it out, perhaps Canon "hit the wall" sooner than
experts have suggested. (Actually probably not, as the pixel pitch for
the 5DII is finer than the 1D II, so at least in the good old days,
Canon's sensor fab was capable of increasing megapixels as well as
improving QE/noise performance).


So: how do you arrive at a QE from the plot, what do you actually
measure and call QE, and how do you get to these numbers?

Anyway, I hope this is convincing evidence that read noise
(at ISO settings where it's relevant)

Nope, the lower the ISO setting, the higher the photon noise
in electrons. Hence: not relevant.

But the higher the # of photons captured and so the lower the shot
noise, the s/n ratio is increased, if read noise is kept about the same.

Reading noise in electrons is *much* more relevant in high ISO: the
electron count multiplication factor is doubled at every ISO stop.
Hence also the reading noise in DN (what Claff calls ADU, I think).

is /much/ higher with the Canon 5D
III (Canon's best FX sensor to date) than the D800,

Nope.

You bandied some numbers. I don't believe in merely your say-so.

and if this is
quantified using Bill Claff's "Photographic" Dynamic Range measure (ie
s/n ratio as would be seen in a real print at any given standard size)

Ah --- nope. Not the s/n ratio. At best the (in his eyes)
usable Dynamic Range. And not at any given standard size,
but at ONE print size-viewing distance equivalent. Without
indication what that may be.

The limiting factor to DR in this case /is/ s/n ratio. That seems to be
a point you've missed somewhere.

And always assuming you didn't buy the D800 for the higher pixel
count, but to only print the same old sizes you printed with 8
or 15 MPix before. Once you want to actually use the higher
pixel count.

then the D800 ****es all over the 5DIII (sorry to be so brutral).

Almost any camera ****es all over any other camera if you
measure one parameter only and choose wisely.


Then something else appears...
The 5DIII plot on the chart "takes a turn" at about ISO 25600, it starts
"improving". Oops - by ISO 102400 it's improved so much that it's got
better than 100% QE, and zero read noise. Now this is quite something,
on a par with finding faster than light neutrinos, Canon have broken the
laws of physics.

Or maybe ... the way they measure the data is broken.
That's the most likely explanation.
That's a woefully optimistic guess, IMO. You'll cut yourself shaving
with Occam's razor if you're not careful.

Hmmmm... perhaps it's time for a shave with Occam's
razor... - yes, once you cut the bull**** away, Canon are now applying
noise reduction to raw data /before/ writing raw files.

You mean like Nikon does for a long time no, and not only for
extended range ISO values? Google for Nikon "mode 3" and
learn ...

I'm sure that's been googled years ago. Relevance to (non astro)
photography is nil. Astro guys may prefer Canon, or better still use
cryogenic cooled sensors designed for purpose.

And if Canon were applying noise reduction, that would be
visible in the image, wouldn't it? So ... show pix!

They didn't tell anybody that they were doing this,

... just like Nikon ...

and apart from the curious
beginning of a non-linear curve at ISO 25600, there's plenty of evidence
that Canon are "cooking" their high ISO raw files - but actually no
evidence that they aren't also cooking raws at low ISO.

We KNOW Nikon's been cooking RAW ... at low ISO. So what's
your screaming about? It's only OK when Nikon does it?

Here's another Nikon cook:
| This is unfortunate, but it is my experience that all
| Nikon cameras do some digital scaling after the ADC.
http://home.comcast.net/~nikond70/Ge...rimer/Gain.htm

It's probably the same with all bayer sensor cameras, but at least not
to the extent needed for foveon. I haven't looked for data on other
brands of Bayer sensors, but the (gain) adjustment is going to have to
be applied somewhere. High ISO NR (a non-reversible adjustment) is
destructive.


So, compared to the 5DII, what do we see with the "improved" 5DIII. A
small increase in QE, from 33-49% - nice and probably NOTHING to do with
the sensor itself, but improvements to the filter stack.

"small increase". And what do you call the increase of
linear resolution between the 5D3 and the D800 ... positively
tiny? (After all, it's only about half the small increase in
QE ...)
Sure - I call the increase in linear resolution between 5DIII and D800
small, hardly field relevant. Even more so because in tests (like on
DPR) the "achieved" resolution increase is measured (even in a studio
test) at only 14% - only 1/2 theoretical increase. But apart from the
crazy priced $8000 Nikon D3X, with Nikon we're mainly talking about an
increase from 12mp to 36mp - and that's possibly really worth having (if
you print large).

High ISO seems
a bit improved, BUT they are cooking raw files.

Nikon does too.

Not "high ISO NR". Cynically, I expect all makers are probably going to
start doing the same. Bill Claff "picked up on it", but most people
probably couldn't care less, most reviewers wouldn't notice - except
they'd possibly comment about how high ISO performance had been "improved".

There's practically NO
difference in read noise from 5DII to 5DIII,

Source?
The sensorgen site provides this info. Bill Claff's charts show it
graphically.

If this doesn't tell you that Canon have hit the wall quite a few years
ago with their sensor fabrication technology, then I have a bridge to
sell you.

If the QE increase alone doesn't tell you Canon could easily go
to 31 Mpix and get the same quality as the 5D2, I have a asteroid
to sell you, with lots of silicon --- which they really need for
computer and smartphone chips!

Still capable of taking nice photos - base ISO dynamic range of the
5DIII is about the same as a 2007 model Dx sensor Nikon D90 which are
pretty good little cameras - you can buy a used one for about $350.

Yep, if ISO dynamic range at ISO 200 is the only thing that
matters to you. As I said, choose a single stat wisely, **** on
other camera.

It's not the only thing that matters, if the D800 (or 5DIII) was a lousy
camera with a good sensor, then it's lipstick on a pig. Nikon's D3200,
with pretty good 24mp dx sensor performance, but dark little pentamirror
finder and no AF micro-adjust is the current standard in porcine
cosmetic makeovers, but that's just my opinion.

So why does the D800 have 36 MPix, when a 12 MPix D90 will do?

Because people want megapixels apparently, for real or imagined needs.
I seldom /need/ more than about 6. But I do sell some prints taken with
12mp over the past few years, nobody complains about resolution or
technical quality in any way, but I'd like some more base ISO DR and
some more pixels to make my life easier. I'm not buying a D800. I
might consider a D600, but the US$900 premium (actually $1100 at present
"street prices") over the dx version of basically the same camera
doesn't have me thinking it's the bargain of the century, even if it
seems to be very good value for an FX DSLR.
*1
How much QE is effectively lost by one of the other little quirks of
digital sensors when used with large apertures, as uncovered by DXO he
http://www.dxomark.com/index.php/Pub...s/F-stop-blues

Tell me how they measured that, when the camera does hide it.

I guess in the same way they measure the anomaly between set and real
ISO. If they tweaked the shutter speed instead of ISO, perhaps nobody
would have noticed, although when it's getting up to EV 1, at slower
shutter speeds you'd perhaps hear that something was amiss.

On Fri, 28 Sep 2012 10:11:58 +1200, Me wrote:

Why not use
http://home.comcast.net/~nikond70/Ch...,D800,Ideal_FX
I would have done, but the link doesn't work correctly...

It does for me.
--

Regards,

Eric Stevens

On 28/09/2012 12:37 p.m., Eric Stevens wrote:
On Fri, 28 Sep 2012 10:11:58 +1200, Me wrote:

Why not use
http://home.comcast.net/~nikond70/Ch...,D800,Ideal_FX
I would have done, but the link doesn't work correctly...

It does for me.

OK - that may be a google chrome thing - it worked for me just then when
I tried it with firefox, but doesn't bring up the
"EOS%205D%20Mark%20III" plot on the chart with Chrome.

Me wrote:
On 28/09/2012 5:54 a.m., Wolfgang Weisselberg wrote:
Me wrote:
On 25/09/2012 6:28 a.m., Wolfgang Weisselberg wrote:

So ... your assumption that the problem is read noise is busted.

Hmmm....
So, /minimum/ read noise of the 5DIII is actually even better 2.4 e-

We can assume that Canon managed to at least hold the read noise
of the 5DII in the 5DIII.

Assumptions can lead to wrong conclusions.

Yep.
Like your 'Canon have "hit the wall" WRT this.' assumption.
Or the "Nikon doesn't cook it's RAW files" assumption.
Or the "clipping low RAW values to zero is a good idea" assumption.


There's practically no
difference, but seeing as you mentioned it base ISO read noise is
actually very slightly worse for the 5DIII (compared to the 5DII).

That's true from your point of view, since you believe the data
gathered and graphed by DxOmark and interpreted by a third party.

At least some numbers given by DxOmark have too many digits, I
fear they can't be that accurate. *And* their way of measuring is
.... ah, 'unique', their results don't match with other sources.[1]
Thus I don't put that much faith into DxOmark and the rest of
their numbers, nor with what other people try to derive from
their graphs.


Sorry, 5DIII read noise measured at 2.4 e- is at (Set) ISO 12,800 (=
real ISO 10,084 - Canon don't tell the gospel truth about ISO either -
all darned camera makers are prone to exaggerate it seems).

I call bull.
1. http://www.dpreview.com/reviews/cano...5d-mark-iii/18
(And they don't find that with all cameras.)
2. ISO 10,084! Not 10,083 and not 10,085 and certainly not
10,000.
The difference between 10,000 and 10,084 is that 10,000 is
0,82% lower. And 10,083 is less than 0,01% lower.

How are you to properly measure that to this exactness?
And --- how comes they report 1/3rd stop difference when
dpreview finds it's correct within 1/6th stop?

http://www.sensorgen.info/CanonEOS_5D_MkIII.html
DPReview is hardly a reliable source.

I see:
1. an attack upon a source that report differently (and which
DxOmark --- the source of sensorgen's data --- regards as
reliable enough to partner with)
2. no answer to point 2. Interesting.

BTW: Sensorgen.info is not even a primary source, just a site that
reinterprets DxOmark data. In addition, the sources sensorgen
cites for it's numbers at least had the grace to give error ranges
of their curve fittings --- which vary by a factor of up to 17
between a limited number of cameras.
Sensorgen seems to think that such stuff just confuses it's
readers ...


Compared to
shot noise at this sensitivity setting this read noise is like farting
against thunder.

So ... reading noise no problem? Fine! I'll come back to that!

Base ISO 100 (set on camera) read noise is:
5D III 33.1 e-
D800 2.7 e-
It's a very very big difference.

2.7 e¯ at ISO 100 for the D800? Souce, please!
What I see is
http://home.comcast.net/~NikonD70/Ge...Read_Noise.htm
4.7 e¯ already at ISO 400, and --- as always --- reading
noise in electrons increase as the ISO goes down and more
electrons are counted. Sometimes faster, sometimes slower.

And 33.1 e¯ for the 5D3? Source please!
http://www.clarkvision.com/articles/...mance.summary/
reports 30.1 and 23.5 for the 5D and 5D2 respectively at ISO
100.

http://www.sensorgen.info/
Browse data to your heart's content.

See above: it's not a reliable sources, since it's not a
source at all.

The figures won't tie up exactly
with Roger N Clark's, but they're close enough

Please list all the values of Roger side by side with the
same values of sensorgen.info, then we'll see about "close
enough".

Just as a hint:
Roger sensorgen %
5D ISO 100 30.1 31.5 4.7%
5DII ISO 100 23.5 27.8 18%

5% *may* be difference between different temperatures and
measurements. Or the offset factor in sensorgen's data. But THEN
18%, when we have already established a base line? Really?
Is that what you call "close enough"? Being off by a factor of
at least 4 between 2 cameras?

Pardon me, something's broken there!


to see that there hasn't
been a trend of improvement at sensor level from Canon for years. (Apart
from QE, which as I said - and it was mainly a guess - probably relates
to improvements in the filter stack)

QE is *the* single most important sensor stat.


"Read noise is worse at lower ISOs" is very true for Canon, and very
much less so for Nikon/Sony Exmor.

Yet how much does it matter? You're already saying that shot
noise trumps read noise at ISO 12,800. Now, ISO 100 is 7
stops slower, meaning 2^7 = 128 electrons where there was
only one before.

2.5 e¯ * 128 = 320 e¯ (5D Mark 2)

It also means SQRT(128) = 11.3... times more photon noise
(which is what causes shot noise.)

2.5e¯ * SQRT(128) = 28.28e¯ (5D Mark 2)

Compare that to the 23.5 electrons read noise of the 5D2(!)
at ISO 100. The same shot noise that --- according to you ---
trumps read noise at ISO 12,800 trumps it even more at ISO 100.

I don't see why Canon should be unable to use smaller pixels
here.

No comment?


You posted a link above to Bill Claff's site. His "photographic dynamic
range" is explained he
http://home.comcast.net/~nikond70/Ge...amic_Range.htm
I think that's a pretty good method from which he presents a chart
application, for example comparing the Nikon D800 and Canon 5D III he
http://i48.tinypic.com/fa55yd.png

Why not use
http://home.comcast.net/~nikond70/Ch...,D800,Ideal_FX
I would have done, but the link doesn't work correctly...

It works correctly. Maybe you need to check your browser
extensions.


Just for interest sake, I included a plot for "Ideal FX" - which is the
perfect 36x24mm sensor, with assumed 100% quantum efficiency (QE) and no
read noise at all.
The 5DIII has improved in (*1) QE over the 5DII, from 33% to 49%, Nikon
D800 is 56%.

And the QE of the 1D Mark II sensor (vintage 2004!) is already 38%.
http://www.clarkvision.com/articles/....qe/index.html
Sensorgen has it at 30%.

Sensorgen, as they describe on their site, is
a) providing only a 'light shines on whole sensor,
including pixel gaps, bayer sensor, AA plate, etc' "QE"
b) is at best only giving a relative QE, and needs a corrective
factor multiplied to it.

Which says very little about the silicon ...

You're telling me a 2008 sensor from Canon has less QE than a
2004 sensor from the same company?

No - it's 10% better (33%) using the same measurement methodology,
but seeing as you point it out, perhaps Canon "hit the wall" sooner than
experts have suggested.

Who would these experts be, then?


(Actually probably not, as the pixel pitch for
the 5DII is finer than the 1D II, so at least in the good old days,
Canon's sensor fab was capable of increasing megapixels as well as
improving QE/noise performance).

The silicon's QE is completely independent from pixel size and
megapixel count (and even from Bayer filters). Even if you look
at the whole sensor with pixel gaps and microlenses and AA plate
and bayer sensor and all it's practically independent for all
common DSLR pixel sizes.


So: how do you arrive at a QE from the plot, what do you actually
measure and call QE, and how do you get to these numbers?

No answer?


Anyway, I hope this is convincing evidence that read noise
(at ISO settings where it's relevant)

Nope, the lower the ISO setting, the higher the photon noise
in electrons. Hence: not relevant.

But the higher the # of photons captured and so the lower the shot
noise,

The shot noise (for a properly exposed shot) is *higher* in e¯,
same as the read noise --- in fact, the shot noise rises faster
as you lower the ISO.

the s/n ratio is increased, if read noise is kept about the same.

Yes, you *can* increase the SNR if you lower the read noise.

Tell me: If you have zero read noise and only photon noise, does
that make your photos any better? If not, tell me why roughly
doubling the read noise in DN for every ISO step is better.

Contrast and compa Read noise in DN, relative to their ISO
100 values --- based on sensorgen data (which is doubtful,
especially the 102400 value should be 80-90).

ISO Read noise Read noise
factor D800 factor 5D3
100 1 1
200 2.7 1.1
400 4.9 1.3

800 8.6 1.5
1600 16.0 1.9
3200 32.2 2.9

6400 54.8 5.7
12800 143.0 9.1
25600 298.5 20.4

51200 n/a 43.4
102400 n/a 128.4


Reading noise in electrons is *much* more relevant in high ISO: the
electron count multiplication factor is doubled at every ISO stop.
Hence also the reading noise in DN (what Claff calls ADU, I think).

Well?

is /much/ higher with the Canon 5D
III (Canon's best FX sensor to date) than the D800,

Nope.

You bandied some numbers. I don't believe in merely your say-so.

and if this is
quantified using Bill Claff's "Photographic" Dynamic Range measure (ie
s/n ratio as would be seen in a real print at any given standard size)

Ah --- nope. Not the s/n ratio. At best the (in his eyes)
usable Dynamic Range. And not at any given standard size,
but at ONE print size-viewing distance equivalent. Without
indication what that may be.

The limiting factor to DR in this case /is/ s/n ratio. That seems to be
a point you've missed somewhere.

The limiting factor to DN in a print is the paper and the ink.
If you get 6 stops, you're doing great. That seems to be a point
you've missed somewhere. (And how much can you compress the
DR until it looks bad? He
http://www.clarkvision.com/articles/...nge/index.html
is a 11-stop DR image. As a documentation that's probably fine,
but as a photography? BTW, that was a 10D ...)

The second limit is that you don't buy a 36MPix camera to print
*at the same size* as with a 12 or 18 MPix camera *at the same
viewing distance*. You buy it to print REALLY LARGE and to get
REALLY CLOSE to the print without seeing pixels. That seems to
be another point you missed somewhere.


And always assuming you didn't buy the D800 for the higher pixel
count, but to only print the same old sizes you printed with 8
or 15 MPix before. Once you want to actually use the higher
pixel count.

.... even though I pointed that second limit out right here!

then the D800 ****es all over the 5DIII (sorry to be so brutral).

Almost any camera ****es all over any other camera if you
measure one parameter only and choose wisely.

Shall we compare repeatable PDAF accuracy with the newest lens
models each on the 5D3 and the D800? The D800 is not bad at all
--- but the 5D3 ****es all over it ...
.... and with the higher pixel count, the D800 needs accurate AF
much more for the pixel peepers and those who bought it for the
extra pixels.


Then something else appears...
The 5DIII plot on the chart "takes a turn" at about ISO 25600, it starts
"improving". Oops - by ISO 102400 it's improved so much that it's got
better than 100% QE, and zero read noise. Now this is quite something,
on a par with finding faster than light neutrinos, Canon have broken the
laws of physics.

Or maybe ... the way they measure the data is broken.
That's the most likely explanation.
That's a woefully optimistic guess, IMO. You'll cut yourself shaving
with Occam's razor if you're not careful.

Yep, a conspiracy is more fun, therefore it must be true.


Hmmmm... perhaps it's time for a shave with Occam's
razor... - yes, once you cut the bull**** away, Canon are now applying
noise reduction to raw data /before/ writing raw files.

You mean like Nikon does for a long time no, and not only for
extended range ISO values? Google for Nikon "mode 3" and
learn ...

I'm sure that's been googled years ago. Relevance to (non astro)
photography is nil.

What about those who shoot at base ISO, ~f/8 (on FF, for DOF and
sharpness) and hence very often need long exposures when not in
full daylight and a tripod? Is smoothing the RAW a good idea
for those people who want maximum sharpness and print *large*?

Maybe you're limited in your imagination what 'photography'
can be ...


Astro guys may prefer Canon, or better still use
cryogenic cooled sensors designed for purpose.

Astro guys can have a 60Da (and before than a 20Da), because not
everyone is rich enough for launching cryogenic cooled sensors
into space.

What's Nikon's offering?


And if Canon were applying noise reduction, that would be
visible in the image, wouldn't it? So ... show pix!

They didn't tell anybody that they were doing this,

... just like Nikon ...

and apart from the curious
beginning of a non-linear curve at ISO 25600, there's plenty of evidence
that Canon are "cooking" their high ISO raw files - but actually no
evidence that they aren't also cooking raws at low ISO.

We KNOW Nikon's been cooking RAW ... at low ISO. So what's
your screaming about? It's only OK when Nikon does it?

Here's another Nikon cook:
| This is unfortunate, but it is my experience that all
| Nikon cameras do some digital scaling after the ADC.
http://home.comcast.net/~nikond70/Ge...rimer/Gain.htm

It's probably the same with all bayer sensor cameras,

That's an assumption, based on nothing but speculation.

but at least not
to the extent needed for foveon. I haven't looked for data on other
brands of Bayer sensors, but the (gain) adjustment is going to have to
be applied somewhere.

Yep: in the RAW converter during applying the WB setting.

There *might* be a reason to apply different *analog* gain based
on the WB setting before writing the RAW, but digital scaling?
That's just a bad idea with no merit at all! I can scale digital
as well on my computer. And by scaling only once less quantization
errors creep in.

High ISO NR (a non-reversible adjustment) is
destructive.

And perfectly OK on an extended ISO settings --- an ISO setting
which the D800 cannot even come close to. However, median
filtering (as a NR (and not only in high ISO) --- a Nikon
speciality) is also destructive ...


So, compared to the 5DII, what do we see with the "improved" 5DIII. A
small increase in QE, from 33-49% - nice and probably NOTHING to do with
the sensor itself, but improvements to the filter stack.

"small increase". And what do you call the increase of
linear resolution between the 5D3 and the D800 ... positively
tiny? (After all, it's only about half the small increase in
QE ...)
Sure - I call the increase in linear resolution between 5DIII and D800
small, hardly field relevant. Even more so because in tests (like on
DPR) the "achieved" resolution increase is measured (even in a studio
test) at only 14% - only 1/2 theoretical increase. But apart from the
crazy priced $8000 Nikon D3X, with Nikon we're mainly talking about an
increase from 12mp to 36mp - and that's possibly really worth having (if
you print large).

Now, see, you remembered that "one print size" comparisons
are worthless with the D800 ... why didn't you remember that
when you compared DR between cameras?

BTW, a lot of other people disagree with you.


High ISO seems
a bit improved, BUT they are cooking raw files.

Nikon does too.

Not "high ISO NR".

True, they do it as "high, medium *and* low ISO NR".
Not just at ISO 102.400. Which they can't do. Thus your
complaint is sour grapes ...

Cynically, I expect all makers are probably going to
start doing the same. Bill Claff "picked up on it", but most people
probably couldn't care less, most reviewers wouldn't notice - except
they'd possibly comment about how high ISO performance had been "improved".

Yep, camera makers are in a world wide conspiracy to fake
their sensor data.


There's practically NO
difference in read noise from 5DII to 5DIII,

Source?
The sensorgen site provides this info. Bill Claff's charts show it
graphically.

Where again does Bill Claff explain how he evaluates data for
the Canon cameras and why the different methods of collecting the
data are comparable? And where does he mention that the computer
screens he uses for putting light on the sensor is a) evenly enough
lit b) calibrated at least on the gray axis c) similar enough in
these parameters for all the data from "collaborations" (i.e. third
parties using the same method and sending him the data)?

If this doesn't tell you that Canon have hit the wall quite a few years
ago with their sensor fabrication technology, then I have a bridge to
sell you.

If the QE increase alone doesn't tell you Canon could easily go
to 31 Mpix and get the same quality as the 5D2, I have a asteroid
to sell you, with lots of silicon --- which they really need for
computer and smartphone chips!

Still capable of taking nice photos - base ISO dynamic range of the
5DIII is about the same as a 2007 model Dx sensor Nikon D90 which are
pretty good little cameras - you can buy a used one for about $350.

Yep, if ISO dynamic range at ISO 200 is the only thing that
matters to you. As I said, choose a single stat wisely, **** on
other camera.

It's not the only thing that matters, if the D800 (or 5DIII) was a lousy
camera with a good sensor, then it's lipstick on a pig.

Quite attractive to boars!
Me, I don't find pigs attractive, I eat them.

Nikon's D3200,
with pretty good 24mp dx sensor performance, but dark little pentamirror
finder and no AF micro-adjust is the current standard in porcine
cosmetic makeovers, but that's just my opinion.

Yep, that's your opinion.


So why does the D800 have 36 MPix, when a 12 MPix D90 will do?

Because people want megapixels apparently, for real or imagined needs.

When 12 MPix *will* do, how can there be real need?

I seldom /need/ more than about 6. But I do sell some prints taken with
12mp over the past few years, nobody complains about resolution or
technical quality in any way, but I'd like some more base ISO DR and
some more pixels to make my life easier. I'm not buying a D800. I
might consider a D600, but the US$900 premium (actually $1100 at present
"street prices") over the dx version of basically the same camera
doesn't have me thinking it's the bargain of the century, even if it
seems to be very good value for an FX DSLR.

I would wait for close to the end of the century for a "bargain
of the century" in cameras. At worst, you'll get the grandchild
to the nth power of the D800 for a song.

*1
How much QE is effectively lost by one of the other little quirks of
digital sensors when used with large apertures, as uncovered by DXO he
http://www.dxomark.com/index.php/Pub...s/F-stop-blues

Tell me how they measured that, when the camera does hide it.

I guess in the same way they measure the anomaly between set and real
ISO.

There's no anomaly, and even they admit that.
But if they can measure it, so can other people. Who else has
measured that and found the same results?

If they tweaked the shutter speed instead of ISO, perhaps nobody
would have noticed,

Suuure. People have been testing shutter speeds since *way*
back in the chemical aera.

-Wolfgang

[1] e.g. the ISO measurements of DxOmark are based on a free
interpretation which noone else shares.
http://ricohforum.com/phpbb/viewtopi...tart=30#p16548
http://ricohforum.com/phpbb/viewtopi...tart=30#p16621
http://www.dxomark.com/index.php/Pub...-make-it-clear