Michael,
Thanks for your insight.
One reason the D3 is offering selectable 14/12 bit depth might be the increased file sizes produced in 14bits. Remember one of the main target markets for the D3 is PJ's where the trade-off would not always be towards IQ (especially if talking about hard to see differences).
[a href=\"index.php?act=findpost&pid=165254\"][{POST_SNAPBACK}][/a]
The 12-bit versions may be read out a bit faster into the ADC, as they have more banding noise than the 14-bit versions, even when the 14-bit versions are quantized to 12 bits, in the D300 paired images I looked at.
I really think that people sit around in these big corporations and decide how to time things like the introduction of 14 bits; they wait until some other improvement comes along, and then marry the 14-bit depth with it in a camera release, so that everyone attributes the improvements to the extra bits, so that next time they can use more bits to make people feel that a big IQ advantage comes along with it.
Ideally, I think the camera should offer two different speeds of readout, but the option for bit depth in the slower speed should be a separate option, too, as the signal content of the extra 2 bits is infinitessimal in current cameras, with their analog noise levels. There seems to be a current technology wall on read noise levels; the best performers at ISO 100, like the Canon 1D*mk3 series, the Nikon D3, the Fuji S* cameras, the Phase P30, and probably others, all have almost *exactly* the same read noise, relative to maximum signal, at their lowest full-DR ISO (100 usually), about 0.035%.
12 bits is more than enough to handle the usable signal content here, but packaging it in a higher bit depth, with extra bits (despite the fact that they are almost all noise) forces converters to use more precision in processing the images, and converters do a *lot* of math on RAW data, in series. I believe that any useful benefit of the extra bits (beyond 12) at this point in time could be had if the extra bits were just binary "1000" for 16-bit, and "10" for 14-bit. If I made decisions at a company, knowing what I know, and being as honest as I am, I would offer users the option of reducing stored bits to significant bits only, while all the time maintaining 16-bit working precision with pads of "10" and 1000" and even "10000" for high ISOs with higher noise. IOW, the encoding scheme in the RAW file would automatically append a binary "1" and then enough zeros to make each pixel 16 bits, after decompressing the reduced-bit data. For example, for ISO 1600 on a camera that had no special "high ISO" circuitry, where ISO 1600 would have 15x the read noise of ISO 100, I would use only 8 bits of data in the main encoding, but the decoding process would would append "10000000" to it, making it 16-bit as it is decoded. Extra precision would be available (overkill), but that really doesn't slow down any converter that's loading everything as 16-bit anyway, as ACR does; it aligns input data to their internal data by the LSBs, rather than the MSBs. IOW, ACR loads a 12-bit RAW "xxxx xxxx xxxx" as "0000 xxxx xxxx xxxx" rather than "xxxx xxxx xxxx 1000". When you have a 16-bit RAW, and the least 4 significant bits are all noise ("n"), ACR loads them as "xxxx xxxx xxxx nnnn", hence, automatically giving the extra precision to work in. This, I believe, may be the only reason 14-bit or 16-bit RAW data may give more tones in the output (if they are not attributable to other things) with current cameras and their noise limits.
