Already bashing the 7d MkII because of the Samsung NXI?

[dgatwood]

Nikon D810 uncompressed RAW, 12-bit: 55.9 MB
Nikon D810 uncompressed RAW, 14-bit: 73.2 MB

That's a file size increase of ~30% going from 12- to 14-bit.

I wonder why. Mathematically, I'd expect it to be only about a 17% increase. I wonder if they're doing something lame like using 16 bits instead of 14 to save on CPU bit shift overhead....

Edit: 55.9 / 12 * 16 = 74.53 MB, so when you factor in the fact that part of it is a JPEG preview, EXIF data, etc. that doesn't expand, I'd say it's pretty likely that they're really writing it out as 16-bit data.

 

[jrista]

The Samsung does 420MP/Sec.

The Samsung is also saving 12-bit files during continuous shooting vs. the full 14-bits in single shot mode.

Which probably isn't as big an issue as people think it is. First, Canon doesn't get the full benefit of 14-bit RAW. They lose at least two bits to noise at low ISO. The main thing they gain is finer tonality.

At high ISO, it's a wash. The lower saturation point implicitly reduced tonality. If you figure that your more likely to be using a higher ISO than a lower ISO at 15fps, your not going to be gathering enough data to take advantage of 14 bits anyway.

The only time the use of 12-bit is going to be a real issue is when shooting at a high frame rate at ISO 100 and maybe 200.

I think it would be interesting to see Canon offer a lower bit depth option if it meant we could get a higher frame rate in the 7D II. If you shoot at higher ISO when using higher framerate, it's more efficient.

 

[jrista]

The Samsung does 420MP/Sec.

The Samsung is also saving 12-bit files during continuous shooting vs. the full 14-bits in single shot mode.

So what?

You keep telling us that DR (and by extension 12bits vs 14bits) isn't important except to a small number of people so therefore this difference that you've highlighted is also relatively unimportant. Or do you wish to become one of your own DRones and insist that the extra bit-depth and thus DR is all of a sudden important?

Hi,
IMHO, when I buy a camera that have 14-bits RAW, I expect to have 14-bits RAW also when shooting at 15fps, so I'll be very mad if I buy this camera and found this to be true especially the Samsung specification didn't even specified that it's 12-bits RAW when shooting at 15fps...

So now the question is: Does the Samsung NX1 really shoot 15fps only at 12-bits RAW?? If yes and they didn't mentioned it, I think Samsung going to be in big problem when buyer find out because shooting at 15fps was the "WOW" feature and one of the selling point that Samsung keep mention it!

Have a nice day.

I'm curious, though. When you shoot at high FPS, are you also at higher ISO, like 800, 1600, 3200? You aren't going to get the same tonal range at those ISOs, meaning the use of 14 bits isn't much of a benefit, if any at all. Not just on the DR front, but on the tonal range front...you don't have the tones or the color fidelity at high ISO to justify 14-bit data anyway. The use of 12-bit files is a bummer for lower ISO settings, but if your most likely to be shooting at higher ISO anyway, it really isn't that big of a deal. It's more efficient...smaller data files, meaning you can fit more on a memory card. At 15fps, that could be a real bonus.

 

[jrista]

But given that Canon's 1DX doesn't have more than 12 stops of DR and that stops of DR are bits then it even seems pointless for Canon to have 14bits of raw, don't you agree?

It's quite a lot of colour-information in those 2 bits, not just DR.

Would you like to demonstrate how the bottom 2 bits add "a lot of color-information"?

They affect the entire tonal range. The bottom two bits add dynamic range, and they also increase the total number of recordable brightness levels. Shadows are inherently sparse, so while adding two bits can have a significant effect on DR, it has a minimal effect on the darkest tones because they are inherently low SNR. Moving from 12-bit to 14-bit means you go from 4096 levels to 16394 levels. A shadow pixel that used 2 bits worth of levels in the 12-bit image is still going to use 2 bits in the 14-bit image. However a highlight pixel that used 12 bits worth of levels in the 12-bit image is going to use 14 bits worth of levels in the 14-bit image. An 18% gray pixel that used 6 bits worth of levels in the 12-bit image is going to use 7 bits in the 14-bit image.

To put that in terms of actual useful levels of information, an image that fades from black to 18% gray is going to have 8192 tonal levels of gradation in a 14-bit image, while only a mere 2048 tonal levels of gradation in 12-bit image. The gradient is smoother with 14 bits, even in a camera that has so much read noise it cannot realize the full dynamic range potential those bits may allow.

At higher ISO, this isn't as important. Drop one stop from ISO 100 to ISO 200, and you lose half your tonality right there. You still benefit from 14 bits at ISO 200, but not as much. Drop to ISO 400, and having 14 bits is really no different than having 12 bits, it's 4096 "real" levels regardless (amplification ultimately just wastes levels with duplicate information when you have more bits than you need, unless some kind of dithering is applied...I honestly don't know if the ACD units or any other component in modern cameras dither or not.) Beyond ISO 400, there isn't any useful benefit to having more bits.

 

[jrista]

The Samsung is also saving 12-bit files during continuous shooting vs. the full 14-bits in single shot mode.

Looks like the Canon 7D2 is just over half the speed of the Samsung NX1, which is an improvement on the 420MP/sec vs 200MP/sec where the Canon camera is under half the speed of the Samsung NX1.

Not quite, you have to take into account the actual file-sizes.

File sizes are just what's output to the storage card. That's the final output stage of the pipeline.

Internally, the camera needs to process the entire output of the sensor, so the Samsung needs to digest 5Gb/s (or 420MP/sec) of data even if it writes out less.

Generally speaking that (the bolded) is correct, however, from what I gather about Samsung's NX1 design, they actually switch the readout to 12-bit conversion via the ADC. They do that SO THAT they can achieve 15fps.

First, modern sensors always have a border of masked and inactive pixels. The total pixel count that is read off the sensor is 31 million. So, assuming 31mp, at 14-bit, at 15fps:

((31,000,000 * 14) / 8) * 15 = 813,750,000byte/sec

To read out 31 megapixels fifteen times per second, they would actually need a data throughput of 814MB/s minimum. There is additional overhead for metadata and whatnot, so I'd say a safe throughput is 820MB/s. However, if we assume that the data coming off the ADC is actually 12 bits:

((31,000,000 * 12) / 8) * 15 = 697,500,000byte/sec

The Samsung needs about 700MB/s throughput to achieve 15 fps at 12-bit readout. I don't really know if it's that much harder to make a DSP that can process 820MB/s vs. 700MB/s, but either way, the processor is faster than DIGIC 5+, which I calculated as having 250MB/s data throughput per chip (500MB/s total). This is quite a bit faster than the 7D II even needs in order to read images out at 10fps:

((21,000,000 * 14) / 8) * 10 = 367,500,000byte/sec

So the 7D II's dual DIGIC 6 don't even need to work as hard (assuming they aren't doing extra noise reduction work and whatnot, which we know they are) as the dual DIGIC 5+ in the 1D X. If we just assume that the total potential throughput of the 7D II is 400MB/s with overhead, each DIGIC 6 is only processing at 200MB/s, which is less than the DIGIC 5+. I don't know how much additional processing power the DIGIC 6 needs to do all the things it does, but let's assume it's double. That would mean a pair of DIGIC 6's process at 800MB/s, which is still short of the 820MB/s necessary to read out 15 frames per second in full 14-bit quality. (That's probably not even accurate, as it would be an image processing pipeline...the input rate from the ADCs would still probably be at most 200MB/s per DIGIC 6, it's just that the chip probably does more work "per cycle" for lack of a better term, as information moves through the various stages of the pipeline.)

So, it's not surprising to me that Samsung opted to drop the bit depth at high frame rate. It allowed them to get away with DSP(s) (I don't know if it uses two) that operate at a slower overall throughput, which probably saved on cost. If Samsung made the assumption that most 15fps shooting would be done at ISO 400 and up, then the use of 12 bits does not represent a loss in either dynamic range nor tonal levels nor color fidelity, so the savings was probably worth it, even if it costs them a few sales because some photographers need 15fps with full 14-bit data at ISO 100 or 200.

 

[AvTvM]

We'll soon enough get full info on NX-1 IQ and capabilities ... including comparison pics for 12bit and 14bit RAWs as soon as the NX-1 will be out in the hands of reviewers and users.

Does anyone have a quick link to identically shot comparison pictures @ 12 vs. 14 bit RAWs for a given recent Nikon model?

http://www.digitalcameraworld.com/2014/06/05/nef-file-types-and-sizes-what-effect-do-bit-depth-and-compression-have-on-raw-files/

conclusion:

01 8-bit JPEGs store 256 shades per channel. That’s fine, until you start wanting to heavily manipulate your images.

02 12-bit NEFs capture 4096 shades per channel. This gives much smoother tones.

03 14-bit NEFs capture 16,384 shades per channel, but you often won’t be able to see any difference next to 12-bit NEFs.

 

[jrista]

Assuming that analogy is not too complex for you, are you suggesting that there's no difference in file size or data content between a 12-bit and a 14-bit RAW file?

Given that file size is generally proportional to megapixels, I don't see how that helps.

So you are suggesting that there's no difference in file size between a 12-bit and a 14-bit RAW file.

I haven't seen Samsung specs, but Nikon allows a choice between 12- and 14-bit, and the lower bit depth allows a deeper buffer in terms of number of shots.

Nikon D810 uncompressed RAW, 12-bit: 55.9 MB
Nikon D810 uncompressed RAW, 14-bit: 73.2 MB

That's a file size increase of ~30% going from 12- to 14-bit.

You really should stop trying to argue technical details, you succeed only in making yourself appear inept.

The size of the file will be largely dependent on how the data is stored. It is entirely possible that a 12bit file from Samsung's NX-1 will be the same size as a 14bit file - it all comes down to how the data is stored internally within the raw file. For Nikon's NEF files, obviously they store the data differently in the two situations (and if you look at the documentation for NEF on the 'net then it is at least obvious that there are markers for this.) Until there's documentation on what Samsung does, all that you can do is speculate and create FUD about 3rd parties, nothing more as just because Nikon does something doesn't mean there's any guarantee that Samsung does too.

It's not impossible that Samsung is doing higher bit depth processing in their post-ADC pipeline, or saving the data to memory as 14-bit data. Sony BionzX does that...the sensor readout is 14-bit, the processing pipeline is 16-bit, however then, sadly, they save the data to their memory card with lossy compression. They still gain from the higher bit depth processing, but it still would be better if they stored full raw pixels, instead of compressed information.

Given that Samsung explicitly calls out 12-bit data for high speed shooting, I think it's safe to say they are storing the information in 12-bit in the data file. Otherwise, they could very well ADC at 12-bit, then use a 16-bit processing pipeline to perform dithering of the information, and save it back out as 14-bit information that was better than 12-bit, but maybe still not quite as good as an original 14-bit read. If they were doing that, I think they would have specified as much.