Hi,
My experience is that DR is very seldom a major factor in the kind of photography I normally do, landscape mostly. Would I need extreme DR the resort would be HDR photography. I did experiment with HDR but I could more often than not achieve similar results from correct ETR exposures. I do sometime church interiors and the experience still applies.
I essentially agree with Bill on well capacity and photo shot noise being the most important factor.
On the other hand I have seen a lot of real images demonstrating superior DR in the Pentax K5. Guiilermo Luik has developed some quite advanced tools for extending dynamic range and he really demonstrated how well the K5 handles deep underexposures.
Marc McCalmont demonstrated the K5 having excellent shadow detail compared with his P45+. I´d say that is real evidence, in the P45+ vs. K5 case raw images I evaluated myself. That's a bit hard to ignore. In theory shot noise is more important than readout noise but the K5 images by Marc McCalmont and Guillermo Gluik really indicate that the advantages are for real.
I have quite a lot of respect for Imatest, which I'm using myself, but so do I have for DxO data. The measurements are probably correct, but the interpretation may be misleading. I had some discussion with Mark Dubovoy on the issue and I got the impression that lenses and MTF play a significant role in perceived dynamic range. A god lens can simply produce more contrast on small detail than a bad lens. Similarly, larger pixels will have higher contrast because MTF is higher for lower frequencies. But this has nothing to do with CCD vs. CMOS or bit depth.
But, I see a real problem. People like Miles, Lloyd Chambers, Mark Dubovoy, Marc Mccalmont and Guillermo Luik present a lot of evidence in form or real images, some even in "raw". Some of the folks who are most knowledgeable like Emil Martinez don't use MFDB but humble DSLRs.
It is very hard to find any good raw images usable for evaluation on MFDBs, but pretty much on DSLRs.
I'd also add that it is not particularly easy to make good tests. To many variables around. I don't see anything wrong with evaluating test images made by others, especially not with images made by people knowing what they do. Of course the images must be raw. I have done such evaluations:
On Pentax 645D vs. Pentax D3X:
http://echophoto.dnsalias.net/ekr/index.php/photoarticles/51-a-closer-look-at-pentax-645d-image-qualityOn Leica S2 vs. Nikon D3X:
http://echophoto.dnsalias.net/ekr/index.php/photoarticles/38-observations-on-leica-s2-raw-imagesOn Phase One P65+ vs. Canon 1dSIII
http://echophoto.dnsalias.net/ekr/index.php/photoarticles/41-phase-one-images-for-downloadI'm not a Nikon fanboy, but it seems that Nikon D3X used to be the champ in AA-filtered CMOS technology. So for me the issue was more about the effect of AA-filtering and how much can be gained by moving to larger formats.
My finding was that:
- Regarding sharpness the bigger format wins.
- Regarding noise and DR Phase One P65+ is much better than Canon 1DsIII. Pentax 645D is about on par with Nikon D3X and Nikon D3X is better than Leica S2.
This observations are based on the images I had access to. The observations are pretty much consistent with DxO data. As a side note the Canon 1DsIII has issues with noise in shadow detail at low ISO, it has definitively been passed by D3X and Pentax K5. That said Canon keeps noise down at high ISO due to preamp implementation (probable explanation).
It would be very nice if MFDB and DMR owners would post comparison raw images from their CCD-based cameras and modern top of the line CMOS based DSLRs like Nikon D3X and Pentax K5. The only comparisons I have seen not that is the original posting here by Miles Hecker and the P45+ vs. K5 comparison posted by Marc McCalmont. Please feel free to contribute images!
Best regards
Erik
On one hand, the claim that 16 bits has nor real advantage over 14 bits is not only supported by the experiments that Bernard quotes, but by the experimental data reported by the sensor makers themselves: both Kodak and Dalsa consistently report S:N ratios (engineering DR) values less than 8000:1, or about what 13 bits can encode, and both report dark noise levels of around 10 electrons or more, so that in 16 bit output, where there are about enough levels to count each electron, the last three bits or so are counting dark noise electrons.
On the other hand, what EricWHiss says is plausible:
"I measured the DR of the DMR and the 5D with Imatest years back now and while the 5D had more DR under the technical definition of ISO something close to 12 stops and the DMR had only 11.5 stops, once the threshold is changed to something more useful to photographers (Imatest draws several curves) the 5D's DR had dropped to 8 stops or under while the DMR didn't drop much at all..."
DXO has a measure of something like "stops of tonal range" which is significantly lower than its reported DR, and seems to be of more practical relevance.
Raising the minimum "useful" S:N level used to judge useful DR makes the dark noise floor less relevant at full exposure (low ISO speed) so that larger well capacity and thus better ratio of signal to shot noise can shift the results in favor of sensor with larger well capacity. As I have argued before, lowering dark noise below about 5 electrons will improve some engineering specs (and is vital for astronomy and useful for surveillance cameras and cameras with tiny photosites) but it contributes little or nothing to aspects of observed image quality relevant to high end MF or 35mm DSLR usage: controlling shot noise by detecting enough photons becomes the main way forward.