Pushing the Blacks—Nikon D3x: medium format quality in a DSLR?
Related: digital sensor, dynamic range, git, medium format, Nikon D3X, noise, optics, pattern noise, raw file processing
Jan 22, 2009
As I began shooting the new Nikon D3x in December 2008, I immediately noticed something unusual: images from the D3x seemed exceptionally well “anchored”, with dark tones looking pure, matching real life better than I’d ever seen with a DSLR. This came not long after my experience with the Mamiya DL28 medium format system, and my initial reaction was that the D3x behaved more like the DL28 than a DSLR! Numerous field images reinforced this view, even 30 second exposures in deep blue shade: the D3x images were stunningly “clean”, something I’d never seen before in a DSLR.
The diglloyd blog covered this in various ways: Jan 14, Jan 4, Jan 1, Dec 27, Dec 24 and more. Early on, I stated that the D3x offered medium format quality in some ways, and experience shows that to be a fair statement. In fact, the D3x is so “clean” that I rate its dark-area performance as superior to some medium format backs, where I’ve observed streaking. See my review of the Mamiya DL 28 in DAP for details.
The fashion on the web has been to trash-talk the D3x because of its US$8000 list price (Canon’s 1Ds Mark III carries the same list price). Sounds good over a few beers, but it speaks volumes about objectivity. Premium quality comes at a premium price, and that’s why pros will pay US$50K for a medium format system. At US$8K, the D3x is sure to make the vendors of lower-end medium format a bit nervous, since “good enough” works.
Nikon D3x breaks new ground
So what makes the D3x imagery so special? It has a lot to do with the D3x offering the cleanest signal path I’ve yet seen in a DSLR. It translates into a breakthrough (for DSLRs) dynamic range of ~13 stops at low ISO, where 14-bit lossless-compressed files are clearly mandatory to maintain the full image potential.
The Canon EOS 5D Mark II / 1Ds Mark III dynamic range isn’t comparable at low ISOs, coming in at just over 11 stops. Those figures are from DXO; my field shooting results are in close agreement. A two-stop dynamic range difference is huge, and it shows in the D3x images, not just the dry examples on this page, but in real images under widely varying conditions.
The D3x is unequivocally top of the DSLR heap for image quality in the ISO 50-400 range as of January 2009, and not just due to noise/dynamic range. It’s a shot across the bow for medium format, and to Canon as well. Nikon is back with a vengeance after really worrying me.
Noise has quantity and character
Noise is often poorly understood as only a quantity when in fact its character is equally, if not more important; the Canon EOS 5D Mark II and 1Ds Mark III exhibit visible streaking and grid noise in dark areas at low ISOs, but the D3x does not, except perhaps by pushing ISO 800 by six stops or so (yes I’ve done so!).
Noise and area
No comparison of noise is appropriate unless it takes into account the reproduction size; comparing actual pixels from a 12MP camera to a 24MP camera (same sensor size) has no useful application: when making a print or even an image for the web, the end size is the same, but pixel density varies greatly. The (imperfect) analogy would be scanning 35mm film and 645 film, then claiming that the identical but larger in area 645 film has lower grain. It doesn’t of course, but it needs less enlargement for the same size print. The math wizards at DXO explain it technically.
The examples on this page have been downsampled to the resolution of the 12.1MP Nikon D3. This is mathematically defensible (unlike upsampling), and allows direct comparison with crops covering a larger area of the sensor.
For detailed coverage, subscribe to diglloyd’s Advanced Photography (DAP). The full noise article in DAP covers the Nikon D3x, Nikon D3, Canon EOS 5D Mark II and Canon EOS 1Ds Mark III. There, noise behavior is explored in multiple ways, and also includes the RAW files for download (expected publish date no later than Jan 25 2009).
Test scene and processing
The same Zeiss ZF 100mm f/2 Makro-Planar lens was used on both cameras at f/8, with focus matched precisely, Solux lighting was filtered to 4920°K with a B+W KB-6 filter, as measured by a Gossen Color Pro IIIf color meter at the Macbeth card. Shutter speed was identical for both cameras, bracketing was done to cross-check. Gray balance set from a Macbeth Color Checker card and brightness (luminance) was found to be within 2 points (Lab mode) for the brighter gray patches; histograms were in very close agreement between cameras and across ISO values.
The ETTR (expose to the right) approach was checked via bracketing (+1/3 stop). This pushed the blacks up, self-defeating for a test of noise in extremely dark areas, and resulting in an unrecoverable chunk of “blown” white (for all cameras) on the focus grid at lower center. This was no accident; the test scene was deliberately constructed for just such a contrast range after an exhaustive initial evaluation using four DSLRs. The Canon 5D Mark II dark-area results improved with +1/3 stop more exposure (of course!), but even so were still much noisier than the D3x.
RAW files were processed into 16-bit TIF with Neutral in DPP (Canon) and Capture NX2 (Nikon) with all noise reduction and all sharpening disabled. Full details are provide in DAP.
Up to 64TB @ 12500 MB/sec!
Mac or PC.
Ideal for Lightroom, Photoshop, 8K video, data analysis, etc.
Pushing the blacks
The examples below show just how good the D3x is in keeping the “yuck” out of dark areas. DAP includes larger crops from all four cameras for ISOs from 50 through 1600 for this “pushing the blacks” test.
Keep in mind that these examples apply to the dark tones only; by the time mid-tones are reached noise differences disappear because any low-level noise is drowned out by a much stronger “signal”.
The examples were “pushed” using the Photoshop CS4 Highlights/Shadows feature. The amount used is equivalent to 3-4 stops, which is not much more than the light falloff with many lenses shot at wide apertures (~3 stops combined optical and sensor vignetting). Combined with features like Active D-Lighting, Peripheral Lighting correction, etc, it is perfectly realistic to “push” the shadows 2-3 stops as a matter of routine, not to mention high-contrast scenes that require very dark tones in order to record the highlights.
Click any of the crops below to see larger versions and all R/G/B color channels. These examples are from ISO 50, but results are similar at ISO 100/200/400 and beyond. Crops are shown downsampled to Nikon D3 (not D3x) resolution, actual pixels results are even more obvious, but harder to compare directly.
A striking difference in noise is evident in the very dark areas: the 5D Mark II is noisy even at ISO 50. Wedding photographers take note: noise is not an issue in a bride’s white gown, but a groom’s black tuxedo will be much easier to manage with the D3x.
Click an image to see larger crops and R/G/B channels.
The “cleanliness” difference is even more obvious when examining the individual R/G/B channels (from AdobeRGB here). Click on either image for more.
Contrary to claims seen online, the D3x does not “pin” blacks; field shooting proves that there is ample detail in areas that look almost completely black to the eye before a “push”. This particular example has no detail to show, as it was a flat black piece of Gatorfoam, but numerous field shots show that the D3x records outstanding detail into the darkest areas.
Contrast and color are different between the D3x and 5D Mark II in this example, but results have been cross-checked in multiple ways for validity.
ISO 100 and 400 comparison
To show just how clean the D3x signal/noise is, compare the D3x at ISO 400 to the Canon 5D Mark II at ISO 100 below. As above, these crops are shown after the Shadows/Highlights adjustment. Crops compressed at 100% quality in CS4.
- D3x noise at ISO 400 is superior to the 5D Mark II at ISO 100, remembering that these are dark and “pushed” areas;
- while 5D Mark II overall noise appears about the same at ISO 400 as at ISO 100, there is more evident pattern noise at ISO 400 (grid/crosshatching);
- the 5D Mark II jpeg file sizes are 20-30% larger than those from the D3x because they are more random (more noise), which results is a less compressible file;
- the 5D Mark II has more chroma noise, but this can be addressed via noise reduction quite effectively.
To keep bright red in-gamut, the crops below are in the AdobeRGB color space! View them with Safari, or Firefox 3 with the Color Management plugin; otherwise they’ll look slightly washed out/desaturated. However, the noise comparison will still be evident.
While numeric noise ratings serve a useful purpose, they are no substitute for the human eye, which detects patterns not reflected in a numeric noise rating. The D3x and D3 display a film-like graininess which does not draw attention to itself. The 5D Mark II shows streaks and crosshatching.
Shown below is a 5D Mark II ISO 100 crop (with Shadows/Highlights adjustment, downsampled to Nikon D3 resolution). Mouse over to see ISO 400, where crosshatch noise appears, a grid-like effect. You won’t find this on the D3x. Curiously, the effect diminishes at ISO 800 and beyond, as other types of chunkier noise obscure the pattern! The pattern re-appears under the right conditions.
Shown below are the ISO 6400 red channels from the D3x, D3, 5D Mark II, all at Nikon D3 resolution (as processed, not “pushed”). While the Canon 5D Mark II controls noise very well on average, it does so unevenly, which results in streaks (in this example), or a grid pattern (earlier example). Yet on brighter tones (not shown), it does extremely well. The 5DM2 has poor control of low-level electronic noise just above the black level; it has a “dirty” signal path. Other types of noise come into play for brighter tones and, and brightness drowns out the “dirty” baseline noise.
Note well that praise for the Nikon D3x should not be taken as criticism of the Canon 5D Mark II, the latter offering tremendous value and possibly a better choice (in terms of noise and color) than the D3x for high ISOs. The reasons for this are discussed in the full noise coverage report in DAP, expected publish date no later than Jan 25, 2009.
Especially at low ISO, the Nikon D3x offers the cleanest signal path yet seen in a DSLR, as good or even better than some medium format systems I’ve previously tested. That is not to say that the D3x matches the best medium format systems—it doesn’t—but it has closed the gap considerably in the key area of dynamic range and signal-to-noise in dark areas. And at higher ISOs (800), it easily outperforms some medium format systems, which can look downright “nasty”.
To be clear, noise is more than a measurable numeric value: the character of the noise matters even more, and here the D3x also shines with a near absence of any pattern noise (grid or streaking). Noise numbers do not tell the whole story, but real images do, and I can attest to the near absence of pattern noise with the D3x at any ISO, based on numerous field shots.
Photographers looking for shadow detail at low ISO will find the D3x outstanding due to its ultra-clean dark tones: underexposed or dark images can routinely be “opened up” with amazing results never before possible in a DSLR. ISOs in the 50-200 range are ideal for such work, with ISO 80 likely the optimal choice (ISO 50 drops about 2/3 stop of highlight dynamic range, with indicated ISO 50 (”Lo-1”) really about ISO 80).
Black and white photographers have a new tool at their disposal: the unprecedented dynamic range and ultra-clean dark tones of the D3x open up new frontiers in digital imagery in a DSLR.
According to the DXO tests, the D3x loses its dynamic range advantage over the Canon 5D Mark II by about ISO 600, which is not to say that its quality ever suffers, only that the 5DM2 gains a very slight advantage as the ISO rises, with some risk of pattern noise in dark areas. The non-linear dynamic range behavior of the 5D Mark II is odd, but might be explained by its “dirty” baseline noise in combination with signal processing at higher ISOs. At any rate, the dynamic range difference is small at those higher ISOs, about 1/2 stop.
Up to 65% better pricing than Apple
Lloyd recommends 32GB RDIMM modules for most users (more expensive LRDIMMS are for 512GB or more).
The findings discussed here are consistent with the D3x tests at DXO for dynamic range and signal-to-noise in dark areas, as well as the ISO 100/400 differences shown.
Liked this article?
If you buy a D3x or 5D Mark II or other equipment, please consider using the links below—thank you!