Data updated upto mttran's. Thank you everyone for the contribution.

I'm very glad that the data are pretty consistent although measured by different individuals.

One thing to notice.
I have been suspecting that 5D and 1D-2 are sharing the sensors with the same technology. So are 1Ds-2 and 1D-3. What I mean by "Same technology" is that the two sensors use the identical design and process except that one just have larger die ( more cells ) than the other. The DR data supports the theory.

Another thing:
Canon has been managing to reduce the pixel pitch ( to add more pixels ) while improving the DR ( or noise ) UNTIL 1Ds-3. Looks like they finally ran out of magic. It seems that they have reached the maximum entitlement of 35mm sensors. We will get more pixels only at the expense of the quality of each pixel.

Pondria wrote:

One thing to notice.
I have been suspecting that 5D and 1D-2 are sharing the sensors with the same technology. So are 1Ds-2 and 1D-3. What I mean by "Same technology" is that the two sensors use the identical design and process except that one just have larger die ( more cells ) than the other. The DR data supports the theory.


While the pixel pitches are similar, the 5D pixels have substantially larger efficiency than the 1D2 (4.1 vs 3.3 electrons/raw level at ISO 400); similarly the 1D3 has substantially larger efficiency than the 1Ds2 (5.1 vs 3.3 in the same units). I don't think they are the same technology.

Where do these electron numbers come from?

Pondria - I think you're using the numbers the best way possible - not as absolutes, but as relative comparisons and as some type of general trend analysis. It is disappointing to see the 1Ds3 at 9 stops. Hopefully others can add to the dataset to either validate or augment Stan's results.

In regards to the 40D's 9 2/3 stops, you'd probably want to know if that was with or with out highlight tone priority.

stanj wrote:
Where do these electron numbers come from?


http://theory.uchicago.edu/~ejm/pix/20d/posts/tests/D300_40D_tests/

Table 7. Most of the data here is taken directly from Roger Clark's website; 1D3 measurement is my own. There seems to be a clerical error in Clark's 1D3 reported measurement, it is off by one stop (ISO 100 gain is actually the ISO 200 gain, 200 gain is actually the 400 gain, etc).

ejmartin wrote:
http://theory.uchicago.edu/~ejm/pix/20d/posts/tests/D300_40D_tests/

Table 7. Most of the data here is taken directly from Roger Clark's website; 1D3 measurement is my own. There seems to be a clerical error in Clark's 1D3 reported measurement, it is off by one stop (ISO 100 gain is actually the ISO 200 gain, 200 gain is actually the 400 gain, etc).

A fascinating and informative read, I enjoyed the report.

Me_XMan wrote:
40D has the same DR as 1DsMK2
Very interesting.

I measured / tested 3x, same result. If my method was flawed then it was flawed for both cameras in the same way...

John Black wrote:
Pondria - I think you're using the numbers the best way possible - not as absolutes, but as relative comparisons and as some type of general trend analysis. It is disappointing to see the 1Ds3 at 9 stops. Hopefully others can add to the dataset to either validate or augment Stan's results.

In regards to the 40D's 9 2/3 stops, you'd probably want to know if that was with or with out highlight tone priority.

Without. That mode means absolutely nothing to me, so I didn't use it. Same with the 1Ds3.

I'd also like to see more values for the 1Ds3, as I find the number by itself disappointing; then again I am not disappointed with the actual image quality, and I feel that it's easier to recover highlights with the 1Ds3 than with the 1Ds2. Naturally the devil may lie in the shadows.

While my absolute value of 9 for 1Ds3 may be off, given that I measured three cameras (40D, 1Ds2, 1Ds3) at the same time using the same method, and repeated the test a total of 3 times, always with same results (I even sent the raws to Pondria for verification), I think it's safe to say that _my_ 40D has more DR as measured by Pondria than _my_ 1Ds3. I still prefer the 1Ds3 over my 40D, though

ejmartin, please forgive me if I'm being a bit dense (it happens more frequently these days ) but how is measuring DR on a per image area basis any different from a per pixel basis ? Wouldn't you be scaling both noise and signal by the same amount and hence getting the same DR result ? Or am I talking about the wrong quantity, being a signal to noise ratio instead of an absolute electron count ?

- Alan

I'm also a bit surprised and disappointed to see the 1Ds3 result lower than the 1Ds2, 1D3 and 40D results. Can anyone (Stan ?) confirm what these numbers mean anything in practice - i.e. does the 1Ds3 seem to have less DR than the 1D3 or 1Ds2 ? If it did then I'd expect it to handle high iso less cleanly. If it seems to be cleaner or at least as clean as the 1D3 then it would seem there is a discrepency in the test results (despite being re-tested several times). Maybe it is mathematically noisier afterall but the noise is visibly finer and less intrusive because there are more pixels per image ?

- Alan

jkurkjia wrote:
GeneO wrote:

I do believe they have different amplifiers optimized for different ISO.
Gene



Gene, wow, where did you come up with that information? Has Canon published anything about this subject?

It's certainly possible to switch between multiple amplifiers as long as power and space don't kill you; I guess this approach can be made to work if the switches and amplifiers are integrated into one chip (otherwise the wiring and pads for interconnects would really eat into available space). Oh well, enough speculation on my part, I'll wait to hear back from you regarding the source of the information, thanks.

Joe


Actually I thought I had read that from a reliable source wrt noise and intermediate ISO increments. In any event I lost the reference. Yes, it probably doesn't not make sense.

Alan321 wrote:
ejmartin, please forgive me if I'm being a bit dense (it happens more frequently these days ) but how is measuring DR on a per image area basis any different from a per pixel basis ? Wouldn't you be scaling both noise and signal by the same amount and hence getting the same DR result ? Or am I talking about the wrong quantity, being a signal to noise ratio instead of an absolute electron count ?

- Alan

You are right that DR and S/N ratio are two different things. DR is the ratio of maximum signal to minimum noise, which is the noise with zero signal; S/N ratio is the ratio of the signal and the noise attendant to that signal. Since the noise increases with signal, these are two different things.

The answer to how things scale: Signal (number of photoelectrons) increases proportional to the area, and thus to the number of pixels in a patch of sensor. The noise in each pixel is independent from other pixels to a very good approximation, and so combines as the square root of the sum of the squares of the noises of individual pixels. If each pixel has the same noise (true for a uniformly illuminated patch), the noise will increase as the square root of the number of pixels combined. So both the DR and the S/N ratio go up as the square root of the number of combined pixels. If comparing sensors of different sized pixels, the per area DR is the per pixel DR scaled by the corresponding linear pixel dimension.


Alan321 wrote:
I'm also a bit surprised and disappointed to see the 1Ds3 result lower than the 1Ds2, 1D3 and 40D results. Can anyone (Stan ?) confirm what these numbers mean anything in practice - i.e. does the 1Ds3 seem to have less DR than the 1D3 or 1Ds2 ? If it did then I'd expect it to handle high iso less cleanly. If it seems to be cleaner or at least as clean as the 1D3 then it would seem there is a discrepency in the test results (despite being re-tested several times). Maybe it is mathematically noisier afterall but the noise is visibly finer and less intrusive because there are more pixels per image ?

- Alan

The answer to this follows from the previous reasoning. The 1Ds3 pixels are smaller; if you look on a per area basis, the 1D3 and the 1Ds3 have the same DR.

And this is the proper comparison. When you look at a print, you take it in as a whole, and you look at the objects in it which are a fixed percentage of the frame regardless of how many pixels they contain, so the appropriate figure of merit is to refer DR, noise, etc, to a percentage of the frame, or if considering a given sensor format, to a per area basis.

Alan321 wrote:
I'm also a bit surprised and disappointed to see the 1Ds3 result lower than the 1Ds2, 1D3 and 40D results. Can anyone (Stan ?) confirm what these numbers mean anything in practice - i.e. does the 1Ds3 seem to have less DR than the 1D3 or 1Ds2 ? If it did then I'd expect it to handle high iso less cleanly. If it seems to be cleaner or at least as clean as the 1D3 then it would seem there is a discrepency in the test results (despite being re-tested several times).

Without any scientific tests - just after tens of thousands of photos, I can say:
- The 1Ds3 is noticeably cleaner than the 1Ds2, per pixel and overall
- The 1D3 is noticeably per-pixel cleaner than the 1Ds3, by a good margin
- Overall, the 1Ds3 is about the same, or better, than the 1D3 because it has twice as many pixels.

I like the 1Ds3 images better than any previous 1D series or xxD camera I used, both from "cleanliness" and from my experience it's easier to recover highlights.

ejmartin wrote:
The answer to this follows from the previous reasoning. The 1Ds3 pixels are smaller; if you look on a per area basis, the 1D3 and the 1Ds3 have the same DR.

While I do agree that pixel size has something to do with things, this is a huge and dangerous generalization. Just compare the 300D pixel size and DR, compared to the 1Ds2/3. To me, this eternal fear of "more pixels mean worse pictures because of [blah]" is just one of the examples that shows that people are incapable of learning from the past. I think there's no general disagreement that the 1Ds3 offers overall better images than the D30. I know this is extreme, but if you want an example in the same family just look at the 1D2 and 1Ds2, where the 1Ds2 has smaller pixels, and clearly less noise and bigger DR.

stanj wrote:
ejmartin wrote:
The answer to this follows from the previous reasoning. The 1Ds3 pixels are smaller; if you look on a per area basis, the 1D3 and the 1Ds3 have the same DR.

While I do agree that pixel size has something to do with things, this is a huge and dangerous generalization. Just compare the 300D pixel size and DR, compared to the 1Ds2/3. To me, this eternal fear of "more pixels mean worse pictures because of [blah]" is just one of the examples that shows that people are incapable of learning from the past. I think there's no general disagreement that the 1Ds3 offers overall better images than the D30. I know this is extreme, but if you want an example in the same family just look at the 1D2 and 1Ds2, where the 1Ds2 has smaller pixels, and clearly less noise and bigger DR.

Um, I'm not sure what you're trying to say here, or that we are in disagreement. I was pointing out that, even if smaller pixels have less DR, that doesn't mean that the image suffers. The 1Ds3 pixels have less DR than the 1D3 pixels; however, DR per unit area is the same; DR as a percentage of frame is better for the 1Ds3 because a fixed area is a smaller percentage of the total image frame.

ejmartin wrote:
Um, I'm not sure what you're trying to say here, or that we are in disagreement. I was pointing out that, even if smaller pixels have less DR, that doesn't mean that the image suffers. The 1Ds3 pixels have less DR than the 1D3 pixels; however, DR per unit area is the same; DR as a percentage of frame is better for the 1Ds3 because a fixed area is a smaller percentage of the total image frame.

Didn't mean to argue just pointing out that _in general_, DR per unit area is not the same if you compare sensors in general. 1Ds and 1Ds2 have the same sensor size, but _not_ the same per area DR. Your statement may be true for the 1D3 vs. 1Ds3, I don't know and I don't argue because I didn't do the science. But I know the statement is not true for 1D2 and 1Ds2, in no way you turn it. And I would be willing to bet that it's not true for the majority of other comparisons one can make, either.

Unless I totally misunderstood something, of course, which is quite possible

Stan,

I haven't seen measurements of the 1Ds DR, but given that it's one generation older than the 1Ds2 I wouldn't be surprised if it had poorer DR, even on a per pixel basis, since it's one generation older in technological development. It certainly has poorer DR on a per area basis. I think you are misinterpreting my point. I was not saying that DR per area was the same for all sensors; certainly it is not. I was instead responding to the dismay expressed in this thread that the DR per pixel of the 1Ds3 was smaller than the 1D3, by pointing out that DR per area is comparable for that specific example. I was saying that these two bodies have the same level of technology, and that the difference in DR can be entirely accounted for by the difference in pixel size.

I am saying that DR of pixels is one component of IQ, but it is just the starting point. DR per area is DR per pixel scaled by the pixel pitch, and is a more appropriate measure of IQ that properly compares sensors having different size pixels. One can further account for crop factor to compare different formats, by dividing by the crop factor.

Most people focus on DR of pixels, and base judgments of IQ accordingly. But that's not how one looks at prints; in prints one considers image features and so one should account for the differing number of pixels that a feature of the same image shot with two different cameras will comprise.

By IQ you mean sharpness, right? If so, then I think that's where things are getting confused. Dynamic range in this post refers to how much range can be captured before highlights are clipped and before shadows go to pure black.

ej

i see this term "pixel pitch" tossed about so much by so many and a different crowd have different version of what it actually is

Can you please give us idiots a explanation and tell us what it means ?


J

John Black wrote:
By IQ you mean sharpness, right?

IQ has many aspects, sharpness is but one; S/N ratio is another; DR yet another.

jmcfadden wrote:
i see this term "pixel pitch" tossed about so much by so many and a different crowd have different version of what it actually is

Can you please give us idiots a explanation and tell us what it means ?

J

Pixel pitch refers to the distance between pixels. For instance, the 1D3 has 3888 x 2592 pixels on a sensor of size 28.1 x 18.7 mm, so the pixel pitch is 28.1mm/3888 ~ 18.7/2592 ~ 7.2 microns.