Definitive A900/D3X noise/ISO samples

[jcoffin]

"The D3X’s image sensor incorporates 14-bit A/D converters on the chip, delivering fast conversion, high signal-to-noise ratio and low power consumption.
-Sonolta

That certainly seems to indicate that my guess was wrong -- I probably should have read through their marketing material more thoroughly before even forming a guess.

[mdcromer]

David,

I just read your article on the main page.

I did not realize that you shot at 14 bit with the Nikon! This changes everything.

Two things that I have read elsewhere:

1) the 14 bit mode shoots at 1.8 fps instead of 5 (which is probably just fine in studio, but not necessarily for event photography which is where you are more likely to be using high ISO!)

and

2) The 14 bit mode uses an off-chip A/D converter. It does not use the Sony on-chip A/D convertor or (presumably) the Sony on-chip analog amp.

So,

The bottom line is probably that the Sony analog amplifier is not nearly as good as the Nikon off-chip analog amplifier. Not really any surprise there.

Can you do the same test with 12 bit mode and see if the results are any different? If my points above are correct, it seems very likely that the results will be much closer to the Sony.

[douglasf13]

mdcromer, the A/D converter is on the chip, just like the D300. Nikon says so in their literature. That's why the re-sample, "virtual" 14 bit method I mentioned above seems rather credible. There doesn't appear to be an analog out off the chip for the use of off board ADCs.

[mdcromer]

Well perhaps the "resample 4x" times could account for the greatly improved shadow detail on the high ISO. Neat trick if that's what Nikon is doing! It would also explain the drastic slowness of the 14 bit mode.

David, let's see the same test with 12 bits instead of 14. That should shed a lot of light on things. . .

[David Kilpatrick]

I can't repeat the test itself at 12-bit, but I can do a test at 14-bit, 12-bit, 12-bit lossless compressed and 12-bit lossy compessed; and Sony 12-bit uncompressed and cRAW. Maybe just at 6400, to see whether the bit depth and compression has any real effect.

I was not going to mention that the Nikon setting black level to zero amounts to clipping the blacks; I've been shot down too often for mentioning this in the past. Nikon both clips and crushes the blacks and deepest shadow tones. Canon do not clip, but crush heavily. Sony does neither. They seem content to let everything through... along with the noise.

David

[Vidgamer]

....
I was not going to mention that the Nikon setting black level to zero amounts to clipping the blacks; I've been shot down too often for mentioning this in the past. Nikon both clips and crushes the blacks and deepest shadow tones. Canon do not clip, but crush heavily. Sony does neither. They seem content to let everything through... along with the noise.

David

You can easily enough crush your own blacks in your RAW converter or paint program.

What I gained out of this thread is looking at the DxoMark site, the A100 still holds its own compared to the newer cameras.

I remember people talking about the A100 noise and... well, noise is still there. Really, though, there was noise when we used film, right? I mean, as long as I stay at ISO 800 or better, it looks good to me. I have more capability than I ever had, and using different RAW converters, results can vary greatly. I try to use 100 or 200 most of the time, but I don't obsess over 400 or 800 when needed. When I have a problem is with 1600... if you really need that kind of speed, ok, then the A100 looks like doggie-doo. The newer cameras are definitely better there, but I'm surprised at the Dxo results, even so.

I'm not getting an A900 or Nikon-whatever either way.

BTW, and maybe this was on another thread, but someone mentioned Ken Rockwell ... he says not to take his site too seriously. Read his "about" page.

[Vidgamer]

It's not the same as crushing blacks after the fact....Nikon achieves *top notch* noise control with the methods they use and you can't duplicate that by crushing blacks in your raw converter. You can improve the look of your some of your photos but you won't get Nikon noise performance by simply crushing blacks.

I don't know what magic they must use, but it seems like they use Sony sensors, so you'd think they'd be "close" at the hardware level. At the software level, well, then you should be able to deal with the raw. It's clear that Sony still doesn't handle JPEGs as well, but I think that's a separate issue.

A friend was showing me photos from his older Nikon (I forget which), and he complained about the noise in some of his low-light photos. Yes there was noise, but it didn't seem that bad to me. Probably better than my A100, and probably would look fine printed at reasonable (non-poster) sizes. Looking at the raw file, you could see a ton of noise in some black basketball shorts. If anything, these blacks needed to be crushed! I increased the contrast and brightness and fiddled with some other settings in the RAW converter, and to my eyes, it was much better.

At this event, I took a lot of ISO 800 and 1600 photos. The light was extremely low, and the shutter speed ensured that a lot of things were a bit too blurry. The main thing you notice is, of course, how noisy the photos were. Even downsized, if you look at my flickr photos, you can see how grainy they are. But this one that I processed from RAW does look a lot better.


http://www.flickr.com/photos/9644425@N06/2833544721/

Yeah, I crushed the blacks and ran it through noise reduction. I could have left it where you can kind of see the stands and people in the background, but they aren't exactly that interesting to begin with. So, push them into the dark along with the noise.

It's amazing how much detail there is (along with the noise) so if I were that interested in the background, I'd leave it in. Also, I'm convinced that it's not possible to have really good NR with JPEG. The camera has to do quick-and-dirty processing to get that photo onto storage and make room for the next one. I'm sure some processors are better than others, but I realized a long time ago that if I needed to work on low-light photos, I needed to use RAW.

Rockwell is Rockwell....he's always been Ken Rockwell.

I agree with some of what he has to say, but sometimes it seems like he says a lot of contradictory things. I don't know, I just wouldn't use him as a sole source to back up an argument.

I owned an A100 and it was far to noisy at ISO 800 to get usable 8x12's in my home arena. I loved the detail it gave me outdoors and shooting macro but the noise levels were unacceptable for most low light shooting even with a lineup of f2.8 lenses and below. A later firmware upgrade improved things a bit but I had already dumped mine for an A700. I kept the KM5D as my backup camera instead of the A100 because the A100 could not perform well enough in low light.
...

Not bad considering the KM5D is one year older than the A100...

-Sonolta

I doubt the 5D would sell in today's market with its resolution... 6mp? Clearly, adding pixels causes some compromise, but if its a simple tradeoff of detail vs. noise, then I think it's a fair tradeoff for most people.

I'm just surprised at how well the A100 keeps up with the latest Sony cameras, particularly given how much some complained about the noise. Well, a lot of the noise is still there....

[David Kilpatrick]

I have been processing a lot of A900 images today - pictures taken at ISO 100-800, many at 400, many in extreme WB conditions with underexposure and difficult dark subjects. I've been amazed at the robustness of the ISO 400 files espeically, taking brightness boosts to 100+ in ACR, exposure to 1.5X or more, fill light, WB shifts as wide as to 49000K (blue light in aquariums) and down to 2200K (candle-level electric in museum exhibits). I have dealt with extreme contrasts and strong boosts to Vibrance, Saturation and Clarity - and I have retained sharpening fairly often, but brought in colour NR (never any luminance NR). Many of these images are really extreme in exposure conditions, too - 1/20th at 85mm hand-held etc.

So far I have lost just two or three shots out of several dozen (and those were on a travel tripod, it clearly didn't like exposure times around 1/8th and was vibrating at just the wrong time). While I'm sure the D3X and 5DMkII would also return results, they would need very different lenses as I use stabilisation to get many shots which would be considered unsafe/impossible. The lenses used for most of the difficult shots were 50mm f1.4, 12-24mm Sigma, and 28-105mm Minolta (which has produced much better sharpness than either of the new similar range Canon and Nikon lenses I've tried for full frame).

While the plain noise/ISO results do not favour the A900, it remains my choice for the kind of subjects I'm shooting and the often fairly extreme adjustments I make in ACR (turning very dull flat scenes into saleable ones).

In the process, I've found something else - colour temperature/WB is critical for reducing the noise level in Sony A900 blue sky tones. If you zoom in at 100% in ACR on a sky area, with some sharpening to show off the noise, then adjust the WB slider (and the Tint slider too) you will see grainy noise appear and disappear at different K/tint values, often being worse or better with a 200K shift around daylight values - which are hopelessly variable to start with.

David

[aster]

Hi,

Well, I was about to post just before David's post as it was about the A100 and ISO 400 but I'm glad to read more of David's findings while experimenting with the A900 raw files. It's encouraging towards the A900. Thanks David.

As an A100 owner of the moment, I've always been very careful not to exceed ISO 200 even at night conditions and indoors; usually on a tripod and steering clear of moving subjects that are bound to appear as phantoms, though sometimes that produces interesting results too.
Lately I've shooting sequential photos of a conventional construction in full action; workers and visitors moving in all directions and doing certain tasks and so on. Handheld, ISO 200 was not sufficient to freeze them while in action so reluctantly, I went for the ISO 400 for the first time in daylight conditions. I'm glad I did because I finally got close to the results I'd prefer. Carl-Zeiss 16-80mm was snappy and pricise and at 100% I could see none of the noise I actually was expecting at darker shadowy parts.
I'll be using more of ISO 400 and relevant settings in the future...

Yildiz

I don't wish to bore you with the non-aesthetic images of a construction-in-progress but here are just two out of thousands that I shot lately. I'm happy that the workers are no longer blurry...

[Dr. Harout]

As far as I remember the A100 was good enough at 400 ISO. Yildiz are those shots cropped images and how much?
By the look I can say your camera (and yourself, of course) are doing a great job.

[aster]

Hi Dr. Harout,
Thank you for the encouraging comments !
The photos in my original post are only resized to upload to the forum.
Here are the cropped from 100% full-sized in-camera jpegs. The conditions for the shooting were:

I was standing 15-20m away from the subjects at a higher spot; handheld; Continuous Autofocus; ISO 400; the subjects always moving and shifting position; RAW+JPEG.
100% crops from in-camera jpegs saved at quality level 12 but there still is some detail loss from the original jpeg.
Yildiz

[Dr. Harout]

Yildiz, I see. Wonderful combo the A100 and CZ16-80.
Thanks

[aster]

Yildiz, I see. Wonderful combo the A100 and CZ16-80.
Thanks

Yes, they are actually.

Thanks,

Yildiz

[bakubo]

Sonolta, very nice high ISO cheerleader and concert photos. Oh yeah, the football shots are really good too. Those kind of action shots in difficult lighting takes a lot of skill.

[bakubo]

D3x-A900 RAWs compared at base ISO with RawTherapee:

http://forums.dpreview.com/forums/read. ... e=30423957

D3x-A900 RAWs compared at base ISO with SilkyPix:

http://forums.dpreview.com/forums/read. ... e=30424761