D3 vs D3x

Indirectly through Nikon Canada I had a chance to have a D3x over Christmas.  There have been endless criticisms of Nikon’s pricing on the D3x, a whopping $8000 US, or $9450 CDN, and I won’t touch upon that here (much).  What I wanted to analyze for myself was how it compared to the D3 at the same resolution: noise would likely be increased due to the smaller sensor sites, but a downsampled image would reduce noise as the individual noisy pixels would be averaged out.  So how would these two opposite factors work out in the D3x? Would the resampling be able to keep the overall noise close to the D3’s phenomenal quality?  Or would noise increase so quickly as to destroy any smoothing done by the resampling?  Basically, would the D3x fulfill the role of the D3 if pressed to work in high ISO situations.

Just to set the stage, I do a lot of studio model shooting and very often run into the D3’s low ISO limit of 200.  I would love an ISO 100 or 50.  Resolution is important, but so are smooth skin tonal transitions.  On the other hand, I do available light event shooting as well, and high ISO performance is also important.  So I need a combination of a D3 and D3x.

I set up a very basic test bed: a 70-180 macro lens, chosen mainly so that I could minimize the amount of room my test Christmas scene took; as well, its tripod mount allowed me to equalize the focusing and framing when I switched between bodies without having to dismount the body from the tripod, thereby reducing any sharpness issues related to focusing variations.  The macro lens was also stopped down to f/8 to get into the sweet spot for sharpness before diffraction set in; mirror-up and cable release were used.  The images were shot in 14-bit RAW and processed in Capture NX 2 with the same camera settings.  The D3x image was then downsampled using bicubic interpolation and a small amount of sharpening was used to match the D3 sharpening (both images show some haloes around high contrast areas).

The quick test images are below.

Sharpness
The resampled D3x images show a sharpness improvement over the D3 images.  This should come as no surprise as any softening effects of the AA filter and Bayer sensor layout and interpolation are going to be reduced by the resampling.  This is most evident in the text areas where there is high contrast between the background and text characters.  The D3 is showing softer text and in fact has noticeable moire in some of the characters (which means that the focusing was bang-on) while the D3x image clearly has crisper text.  At all ISOs up to 6400 (the maximum on the D3x), the D3x could resolve more detail in the text than the D3, despite similar noise characteristics.  The crispness is noticeable in the 100% view.  Resampling down is good!

Noise
Noise, or the lack thereof, was a surprising discovery.  The D3x produces extremely noise-free images at low ISOs.  At base ISOs (100-D3x, 200-D3), black areas are surprisingly smooth and devoid of noise, even slightly less than the D3.  It almost appears as though the black levels have been clipped in software.  However, a quick check against the D3 image shows that the same subtle shadow details are still there, so this is amazing sensor noise performance for the D3x.

When ISOs are increased, noise increases, of course.  Per-pixel noise at each camera’s native resolutions also seems to be pretty similar up to ISO 400 as well, after which the D3 slowly pulls ahead.  That’s not to say the D3x isn’t good, just that the D3 is so good.  Thankfully, noise appears more film grain-like as with other Nikons, and is less objectionable below about ISO 1600.  Above ISO 1600, the colour noise does pick up and details get lost as well.  Resampling results at 12MP show that up to ISO 800 or even 1600 the images generated by both cameras appear identical, but with the D3x having the edge on detail throughout.

There is an interesting characteristic I’ve noted on some high contrast areas — there is some sort of noiseless dark band or halo around objects at high ISOs (around the model lightbulb), almost some form of processing artifact as I see it appearing after around ISO 400, which rules out the lens being the cause.  I thought it was a D3x thing, but I see it on the D3 as well, so I assume it’s noise reduction.

Dynamic Range
I did not make a detailed DR analysis, other than to investigate highlight headroom on overexposure and subsequent recovery.  In that regard, I don’t see much difference below the ISO 800 or so mark — the D3x provides a good amount of latitude for what I would use it for.

Conclusions
The D3x is an interesting conundrum.  On the one hand, there’s the price.  On the other hand is arguably the most advanced 35mm DSLR today, with stellar image quality and resolution.  Where does it fit in?

The performance of the D3x is amazingly close to the D3.  Below ISO 800, there doesn’t appear to be much difference between the two cameras, with the D3x providing the extra resolution of course, so it would be the better choice.  ISO 1600 could go both ways, and above that is D3 territory of course.  Both cameras deliver extremely similar colour, a very useful characteristic when working with multiple cameras at the same time.

Now, if you resample, it buys you perhaps two stops, so an ISO 3200 image on both cameras at the same resolution appear pretty darned close.  The D3x can be pushed past its ISO 800/1600 comfortable full-resolution limit to its Hi-2 (ISO 6400) limit with good results this way.

So for anybody not needing the stratospheric 12,800 and 25,600 ISOs of the D3 and the high framerates, the D3x could just as easily be a D3 replacement.  It’s that good.  You won’t have to choose between high-ISO performance and high-resolution for just about any shooting situation — you can get it in one camera.  This is if you resample down to the D3’s resolution of course.  It’s almost cheating — the noise goes away and the sharpness increases when you resample.  Above ISO 1600, one gets into uncalibrated ISO land and colour suffers, just as it does on the corresponding Hi-1 and Hi-2 settings on the D3.  But it’s not as bad as you’d expect, manifesting more in increased noise in my case.  Details are still holding up extremely well, especially those not in deep shadows, where the noise lives. 

For studio work, the extra usable resolution coupled with the one lower stop of ISO makes the D3x a phenomenal camera.  The sharpness and clarity are real and the noise performance is outstanding.  There is no way upsampling could replicate the extra resolution.  On the flipside, the extra resolution does make a difference when resampled down, so there is no downside to resolution in this case (other than a little sluggishness when reviewing the massive images).  Although the camera slows to 1.8fps in 14-bit mode, the vast majority of studio lights today are unlikely to recycle this fast anyway, so the limiting factor is not the camera.  Landscape shooters should have no problem.

Personally, the D3x is much more tempting than I expected.  I was prepared to reject the high resolution because I felt I would be give up too much high-ISO capabilities.  Since I’ve had a chance to test the D3x, I’m very impressed and intrigued by the both its raw image quality as well as the noise-free resampled image quality, the latter characteristic a key one for being a D3 replacement.  It certainly could fit the D3 replacement role, especially since I have a D700 as well if I need all-out high ISO performance.  The D3x’s studio performance is what’s most compelling — a noticeable jump in image quality while utilizing my investment in Nikkor lenses is what I’m looking forward to.  The studio / architecture / landscape role is the D3x’s niche.

It all comes down to value.  Is it worth nearly $4000 over the D3?  That’s a terribly tough call.  If the premium were a mere $2000, say, over the D3, I think it would be a done deal.  As such, it’s just so tantalizingly worthwhile an upgrade, yet just that little bit out of the comfort zone.

What remains is to compare this with a medium format camera (hopefully a not-so-far-in-the future test!).