Being new to RAW (I previously shot ONLY .jpg) what are the real world advantages to 14 bit vs 12 bit RAW images? Realistically can you see the difference in 14 vs 12 bit? Raw noob in need of info.
Realistically? None. I'd love to be proved wrong though. I tried taking some side by side shots with my D700 with 12-bit and 14-bit and even when pushing the shadows up by 2 stops I couldn't really saw I saw a difference. Others have claimed smoother tones and less banding with 14-bit but I haven't noticed.
I doubt anyone can PROVE 'daBean wrong - not with the cameras we use today. There's a VERY small difference, and you need to look VERY closely at pictures pushed more than 2stops :-)
It's measurable, but I doubt that it's double-blind-test distinguishable in a real photo unless you're doing one-shot HDR's, and I'm not too certain in that case either. "Lossless" compressed 14 take no more card-space than 12-bit though, so why not? The difference is slightly "larger" at higher ISO's as the main difference is an increase in pixel-read precision (which has nothing to do with 12/14 bits. They could have implemented increased precision on a 12-bit out, and lost very, very little).
It may fill up your buffer a couple of shots shots sooner though - so machinegunning shooters may see some difference in maximum numbers of frames until the fps slows down... I think. Depending on your card & camera :-)
My experience is about the same - I really can't see much difference between 12 and 14. I might be able to "push" my 14 bit files around a little more in post, but I'm not even sure that's the case. I usually shoot 14 anyway, but if I need my D3x to go faster than 1.5FPS, I just to 12 bit with no hesitation.
14-bit can be useful in the case of badly mis-exposed images or when in need of large moves in exposure for DR. But it's generally of very little benefit today. Think of it as a step into the future: eventually, it'll matter and be significant, but today it's not a big deal.
That being said, I use the 14-bit except when I need fast bursts. If it's there, why not? I want the best possible IQ, so any tiny bit is welcome. Not strictly necessary, and I don't blink or hesitate to move to 12-bit when FPS calls... but if it's there, I'll use it.
firewireguy wrote:
The main difference is that you need a proper beast of a machine to process 14 bit images.
I'm not sure it actually make any difference. The computers are going to hold both 12bit and 14bit numbers in a 16bit variable - because that's how computers work. The big difference is processing 8bits vs 12/14/16bits vs 32 bits. Anything between 9 and 16 is going to required the same processing power and memory.
This has been discussed a lot in the past, and there are some good articles around on the web. In theory, there shouldn't be much difference because 12 bits is enough to record the full DR of the sensor.
The D3x is an interesting case because it writes the 14 bit data much slower. I have heard that it is written twice to minimise data read noise. This is the reason D3x has such low noise at low ISO in 14 bit mode (refer to the black frame tests).
Roughly 6 dB of dynamic range per bit. 12-bit = 72 dB dynamic range and 14-bit = 84 dB dynamic range. Every 6 dB corresponds to doubling the voltage sensitivity to the analog-to-digital converter (A/D), so you should see about a 2-stop improvement in light sensitivity (ISO). However, in practice camera manufacturers have increased the pixel density of the sensor, thereby increasing the noise. In reality we've seen a 1-stop improvement by going to more bits in the A/D.
Practically all of the benefit is achieved by using any raw format (12 or 14 bit) instead of 8-bit jpg format. From what I have read there is no benefit to 14 bit for most cameras with a possible exception of the D700/D3 family and perhaps the D3x, but only at minimum ISO where DR is greatest. Other cameras do not have the necessary useful dynamic range to get extra signal above their noise floor. i.e. with 14-bit you're getting an extra 2 bits of noise floor and 0 bits of extra usable data.
This was thrashed to death some time ago in a very long and very technical thread in the Canon forum. A lot of good stuff was published in that thread but it was certainly not for the non-techies. Ah, those were the days, but anyone making a thread like that on this forum would get flamed for sure. Every real advantage espoused for 14-bit had already been achieved by 12-bit/raw.
That won't change in future for any existing files but will no doubt change for new files from new high-end cameras in future - some day. Meanwhile, if you use a camera that suffers significant slow down or loss of buffer space when using 14-bit files then stick with 12-bit files and know that you are still capturing everything that is useful.
I would tend to disagree. As Canon has demonstrated with the 7D sensor, it has improved the quantum efficiencies of the photodiodes; therefore, given a lower bias voltage to the sensor, the analog signal will be less noisy and you can take the benefit of a 14-bit A/D. The ISO sensitivity on the 7D, from initial reports, seems to be a 1 to 2 stop improvement given a very high pixel density.
I would think Nikon, or rather Sony, will improve quantum efficiency as well in the next set of releases.
I can barely see a difference in most work, but if you need the most tonal gradations you can get, then 14bit gives you a better, less bandy looking transition.
I used it a lot at first, but don't much now.
As for the pixel density part of the equation, I agree that the signal received by the sensor is improved in sensitivities by the conversion in 14bit. That probably accounts for the better tonality in colors.
that is not much of a point if it is visible in all images. it is very visible in images where are gradation of single color. in studio and fashion work it is very often difference. we did the test together when we were comparing if lossles compressed format is same as not compressed and we seen in many test scenarios visible difference in prints and on good screen (12-14bit and no difference in lossless compression to not compressed).
would post the pics but am too lazy to look it up and will do nothing to me if you believe it or not...
but question is, when you can, why not? difference in sizes is small, cards are big, PCs are fast so why to artificially limit? you buy expensive body and then not use it at its max. same like buying 2000Eur lens and then stick cheap filter over it.
Alan321 wrote:
From what I have read there is no benefit to 14 bit for most cameras with a possible exception of the D700/D3 family and perhaps the D3x, but only at minimum ISO where DR is greatest. Other cameras do not have the necessary useful dynamic range to get extra signal above their noise floor. i.e. with 14-bit you're getting an extra 2 bits of noise floor and 0 bits of extra usable data.
Differences are fairly negligible, but still visible to me in these examples. These differences become more relevant at higher ISOs rather than lower ISOs because such extreme exposure values do not appear in base ISO images unless you are doing extreme manipulations (these images in fact appeared completely dark until EC was boosted 2 stops, brightness was raised, and an extreme curve was applied).
Thom Hogan has written about this, and he specifically wrote that 14 bits was more useful at higher ISOs. IIRC, he also wrote about the usefulness of 14 bits on the D3x in extracting maximum image quality from it -- Marianne Oleund at DPR also wrote about 14 bits delivering significantly better quality for the D3x, but she does not think the D300 can really take advantage of 14 bits.
unless you are shooting with a D3X, none of Nikon's other bodies show enough improvement to matter. the D3X shows the mosts difference and even then, it's not large unless you deliberately shoot taking advantage of the shadow detail abilities of the camera.
Well, the readout accuracy difference makes more of a difference the deeper into the shadows you go. And every full step of ISO you increase "lift" this problematic region one step further into the picture.
Remember, ISO has nothing to do with the light hitting the sensor, or the readout from the sensor. It is applied after readout, before AD conversion (to12- or 14-bits).
The D3x is the one body right now that has some real substantial performance to gain from using 14-bits, but that has nothing at all to do with the two extra bits, it's all in the readout scheme. I can see the difference in Tony's example, but I can hardly call it "substantial"... :-)
But if you don't need the fps, use it everywhere that you can. Why waste a potential performance increase?
Remember, ISO has nothing to do with the light hitting the sensor, or the readout from the sensor. It is applied after readout, before AD conversion (to12- or 14-bits).
On the contrary. ISO is the light sensitivity of the sensor, that is voltage applied to the sensor for signal gain, in order for the sensor to convert light to voltage. Should the sensor be more efficient in doing this conversion, it would require less voltage, hence a lower ISO setting.
DaveEP wrote:
I'm not sure it actually make any difference. The computers are going to hold both 12bit and 14bit numbers in a 16bit variable - because that's how computers work. The big difference is processing 8bits vs 12/14/16bits vs 32 bits. Anything between 9 and 16 is going to required the same processing power and memory.
"In theory, theory and practice are the same. In practice, they are not." I know my PC is a LOT slower processing 14bit images over 12 bit ones.
Also it completely depends on the way the memory is managed. It almost certainly isn't allocated as simply as that. The memory could be allocated as a three variable struct or an octet array. In which case you could use 5 octets to store 12 bits of data per colour, but you'd need 6 bits to store 14bits per colour. But then if you were using a class to store this data then you'd have extra overhead and you could always run into problems with missing cache lines based on the CPU you're using.
In Windows at least there is only support for 8 bit colour per channel so you would need to perform some interpolation on the images which could well be more costly on a 14 bit image.
