Wednesday, July 1, 2015

Canon EOS 5Ds vs Nikon D810

Left: Canon EOS 5Ds   Right: Nikon D810

There's a phrase that is used every so often around here, and that would be "beyond the scope of this blog". As in "getting the best out of the Canon 5Ds and the Nikon D810 are beyond the scope of this blog". If you know why you need one or the other of these cameras you likely will have decided by now; otherwise, if you have to ask, you probably don't need either. Nonetheless, lets see how the very best stack up against one another.

5Ds Key features

What has changed since the 5D Mark III:

  • New 50.2MP Canon-designed sensor 
  • 150k pixels exposure meter (RGB+IR, like 7DmII
  • Reinforced tripod mount
  • Electronic first curtain shutter option in Live View
  • Dual Digic 6 processors
  • Artificial light flicker reduction (like 7D Mark II)
  • New mirror mechanism (like 7D Mark II)

What hasn't changed:

  • 61-point AF module, 41 cross-type, 5 double-cross type
  • 1080/30p video (7DmII can do 60p)

What's worse:

  • No headphone jack
  • No clean HDMI output
  • Battery life slightly worse

The end result is a mixture of ultra-high resolution feature set combined with features that seem curiously...crippled. The exposure meter is upgraded from the 5D Mark III, which is synergistic with the increased sensor resolution. Though the 5D Mark III autofocus system is good, a (theoretically) better unit now exists in the 7D Mark II. The fact that video hasn't changed at all and is arguably worse than the 5DmIII indicates that Canon is being more single-minded about the EOS 5DS' mission than it was with the general-purpose predecessors. This is a shame; think of all the wedding photographers who need to do video and large group photos, and how much easier it would be if all of that functionality was available in one camera body.

The 5Ds shutter uses a new motor-driven mirror system. A conventional DSLR mirror is sprung to flip in one direction. The 5DS, however, shares the same mirror mechanism as the 7D Mark II, where the reciprocating action of the mirror (up-down) is articulated by a crank motor and connecting rod system. The motion of the assembly resembles the way that the crankshaft and pistons articulate in an automotive engine: The advantage to doing it is way is that the upwards motion of the mirror isn't just a straight linear acceleration followed by an abrupt stop; the motor housing has a brake lever that decelerates the upwards motion to soften the landing.

Fine Detail

Optical Low-pass Filter

Like the D800/D800e, the 5DS is split between the regular version and the "R" version which uses a cancelled-out optical low-pass filter. The reason why the anti-aliasing filter isn't removed completely is because the optical system at the microscopic level has some depth to it; in order to produce both cameras on the same production line, the dimensions of the sensor stack assembly have to be kept the same to maintain manufacturing tolerances.

A more interesting questions is why Canon decided to bifurcate the 5DS line at a time when Nikon consolidated the D800/D800e into the D810. Nikon made the decision that the upside (better resolution) was greater than the downside (moire and false colour patterns)  for the majority of their users. Real-world use shows that unless you are shooting at tight repeating patterns (fabrics), moire is not a significant issue with the D810... and that's with a camera that has a lower pixel density than the 5DS. As pixel density climbs, the ability of the camera to discriminate high frequency detail goes up. The Sony-sourced medium format sensors used on the Pentax 645Z and the Hasselblad H5D-50c don't use anti-aliasing filters either. In this regards, the 5DS is out of step with the rest of the industry. It's a curious move considering that the real-world difference between the D800 and D800e was equivalent to a very modest amount of sharpening in post-processing, and that the D800e itself sold in very few numbers relative to the D800. The answer is likely that the 5Ds kept the large contingent of fashion Canon fashion photographers in mind; for fabrics and textiles, moire is still an issue


The difference between the 50mp 5DS the 36mp Nikon D810 is equal to an 18% increase in linear resolution. For perspective, 10% is roughly the threshold at which people can tell that there is a difference. In qualitative terms, given side by side images, most people can easily say that the 5Ds image is more detailed, but they won't be able to quantify it byeond that.
Though 50mp is a big number, the difference between the 5Ds and the D810 is just under the difference in linear resolution between the Nikon  D810 and the 5D Mark III. To put the increase into perspective, at 300 DPI, you can produce detailed prints with a 5D Mark III at 13" x 19", the largest size that the home-oriented Canon Pixma Pro-100 can accomplish.

Inevitably, the question arises: do you need all of this resolution? The answer is: if you have to ask, then the answer is no. The longer answer is much more nuanced, of course. If you are doing large prints more resolution is a better thing. However, many people don't notice/pay attention to the colour detail lost to the Bayer array. This is why more pixels is always better if there aren't any downsides. If the 5DS can create a more detailed image than the 5D Mark III without a noise or dynamic range penalty, then it is a better sensor by virtue of the fact that it is sample more image data. Put another way; you don't have to use all 50mp to get the benefit of more pixels; the 5DS image down-sampled to 5D Mark III dimension will appear crisper and sharper. You can see this with the Nikon D810; when you down-sample its images, you still get better visual acuity than with smaller sensors at native resolution.

This is a Canadian $20 bill, as shot through the EF 100mmf/2.8L IS USM on the 4Ds and the AF-S 105mm f/2.8 Micro on the D810. Both will produce absurd levels of detail under these conditon.

Canadian $20 Bill

Just a note about the subjective appearance of these images. The Canon has a higher default sharpening level and a steeper tone curve. That can be mitigated by increasing sharpening and contrast on the Nikon, but in this Case, the 5Ds displays virtually no mechanical camera-related vibration during live-view, and gets the utmost detail out of this example.

100% crop
Equalized sharpening and colour

Just as a matter of perspective, if you hold the $20 bill up to your face you can make out these individual lines with the naked eye, but you almost certainly will not be able to see the white dots within each individual line, nor the faint dots on the white background.

Image Noise

The following is an ad hoc demonstration of the camera's JPEG output quality through the ISO range. Naturally, the best use of these cameras is with the RAW files, but the JPEG tuning often gives insight into the mindset of the manufacturers. Both cameras were set to spot meter, high ISO noise reduction off, dynamic range enhancement off, default tone curves. Shot with their respective 50mm lenses at f/5.6

Pay attention to the legibility of the soda bottles for detail retention (or lack thereof) as the ISO value rises, speckling in the broad colour patches for overall image noise, as well as the harshness of the reflections and shadows as dynamic range decreases. These samples are not directly comparable to similar samples found elsewhere in this blog because of variable ambient lighting conditions. Click on images for 100% crop view.

Canon 5Ds ISO 6400 - downsampled

As with above, the Canon will show a subjective advantage in the following examples because of the higher default sharpening level. For a high pixel count camera, the Canon holds up fairly well against the D810. Per pixel noise is a step behind and at each level, but there's enough resolution here to mitigate that for most purposes except for ultra-printing gallery-sized reproductions. Base ISO image quality is splendid for both cameras; you can see the dynamic range advantage of the Nikon sensor in that the shadows don't drop off to black as qui8ckly as with the 5Ds.

A small surprise: the 5Ds meters closely to the D810 in this situation.Typically, Canon cameras have given longer shutter speeds  than their Nikon counterparts for the same camera settings and shooting conditions; not so in this case. Though this is a specific instance, the 5Ds white balance seems to be more aggressive than that on the D810; the Canon pulls the image closest towards an idealized white balance, whereas the Nikon is not perfect, but a tad bit closer to the the actual (ever so slightly slightly warm) colour temperature of the ambient surroundings.

ISO 100
ISO 3200
ISO 6400

Dynamic Range

One thing that is hard to avoid in current internet discussion is how much Canon has lagged behind the Sony sensor cameras (Nikon, Sony, Pentax et al) in terms of DxO scores. There are a lot of truths and half-truths about this, but it basically boils down to:

  • There is a practical difference in dynamic range between the 5D Mark III and the Nikon D810. This shows up especially in post-processing and with contrasty subject matter at low ISO
  • Be that as it may, the market share positions between Canon and Nikon have not changed appreciably over the past five years, so outright image quality is not the sole determinant of many Canon user's criteria for choosing the system. 

Lost in the jingoism of internet discussion is the fact that DxO publishes a composite score. The final "number" is a weighting of many different criteria, and the value of the score can be changed according to how the various components are weighted. The new Canon 5D moves forward with the move to 50mp, but the implication is that it has roughly the same dynamic range as the 5D Mark III. Considering that Canon hasn't announced any major changes in circuit manufacturing, this isn't surprising. If it's a case of more resolution but not as good exposure latitude as the D810, the DxO Mark score might still close up between the two cameras.. Given the past history of Canon and DxO, this alone would be considered a victory, but the specs give a clue to what's going on. The rated ISO range for the 5D Mark III is up to ISO 25,600 for native ISO; the 5DS cameras cap out at ISO 6400, two stops less than their predecessor. To put that into perspective, that's the same native ISO range as the EOS 70D.

The decrease in high ISO noise performance is understandable given the increased pixel density, but one that that is poorly understood is that the lagging low-ISO dynamic range is arises from the fundamental Canon image pipeline architecture. Canon sensors have used analog to digital converters (ADC’s) located outside off the image sensor itself. What is means is that a handful of ADC processors (with the 5D MarkIII, 8 are used) share data among all the columns on the sensor by multiplexing the analog signal out to the ADC board. Minor differences between these analog read-out channels produces patterns in the output signal which may be visible when the signal to noise ratio is low, as in the shadows. Because the analog pathway from sensor to ADC is longer than it is with the Sony chip's, the Canon chips tend to have more image noise at low ISO; the analog signal is vulnerable to picking up stray signals from the surrounding circuitry. The longer the signal stays in analogue form, the more chance of inducing extra noise. The difference tends to narrow at higher ISO.

Conversely, Sony chips use ADC's that are integrated into each pixel column on the sensor, and are run with a chip clock speed that is significantly lower rate than the Canon implementation. The signal fluctuations are still present but because there are thousands of built in ADC's running at lower speed instead of a handful running at high speed the signal variation is low and distributed more evenly across the the whole image readout. This results in less pattern noise, aka "read noise". This also improves the low ISO dynamic range compared to the recent Canon chips. Dynamic range is the difference between the saturation capacity of the light-well and the noise floor; because the Sony chips have such low read noise, the noise floor is lower. This has two implications for still images. The first is the idea of "ISO-less" shooting. The second is that it tends to inflate the DxOMark scores for sensors with on-chip ADC's, as the DxO composite scores are weighted to favour low ISO sensors.

However, even though the Sony method is better for still photography, that doesn't necessarily mean that it is the best architecture overall. With the ADC's located on-chip, heat generation can be an issue with higher frame rates, meaning that additional cooling considerations need to be factored into the design. The implications for continuing with an off-chip ADC go something like this:


  • Lower power use on the sensor; less dark current. Good for video and astrophotography.
  • ADC can be designed independently of the image sensor


  • Additional space taken up in the camera body by ADC mainboard.
  • Still image quality compromised at the benefit of improved video performance.

This doesn't mean that it is a case of either/or, though. The very video-focused Panasonic GH4 uses the on-chip ADC approach. If you look at the current trajectory of sensor design, it's very much in favour of using on-chip ADC's. The situation is similar to the CCD/CMOS "debate"; there basically is none, at least not for stills photography.

The final thought on this: if the 5D Mark III was more camera than you needed, then the narrow-ish dynamic range of these cameras don't matter from a technical stand point. Compared to a Nikon, you have to be more careful with how you shoot and you have less leeway to edit in post processing, but great photographers are using both systems.

The following is with both cameras set to spot-meter a shaft of light falling within the atrium. Note that the D810 default tone curve is somewhat steep into the shadows, whereas the 5Ds tone curve is a bit more "open" in this circumstance. This illustrates a point about dynamic range: so much of it depends on how the data is interpreted by the camera or post-processing software, and the range narrows as you move from RAW to JPEG to print. Shot on their respective 70-200mm f/2.8 lenses at f/2.8 and ISO 100:

Canon 5Ds - Spot Meter (1/2000s)
Nikon D810 - Spot Meter (1/2000s)

Here is what a 100% crop of of the scene looks like when pushed 5EV:

Compared to the stereotypes about noisy high-resolution sensors that was prevalent a few years ago, both cameras hold up well, though the Canon does have more apparent pattern noise in the deep shadows. Here is the same scene with the cameras set to evaluative/matrix meter. If you were looking for the dynamic range advantage of the D810, it's ...sort of....there in the lettering on the yellow lettering of the WIND kiosk sign, which is less washed out than on the Canon sample.

Canon 5Ds - Evaluative Metering (1/250s)
Nikon D810 - Matrix Metering (1/250s)

However, what is more true is that with good exposure, you can get great images from both cameras. JPEG quality is a tricky proposition to go by, as the way the tone curves are a factor. The actual advantage, of course, is in RAW processing. Hope you have a good computer if you plan to do that with either camera....

 Concluding Thoughts

If you think 50 megapixels is excessive, there's another way to look at the 5Ds. Don't think of it as a 50mp Bayer-type sensor, think of it as being the equivalent of a 12mp Foveon-type sensor. If you downsample the native 5Ds files to 12mp, you will restore the the colour detail lost to the Bayer array at each of the newly created pixel sites. In other words, there is more to having more pixels than merely having more size and crop space; the extra data also produces smoother tonal transitions and be be used for greater colour-specific detail. We're well past the early days of digital when more pixels meant a trade-off in image-noise and dynamic range; with each new generation, we have gotten more sensor tech, and each time that has also meant that we have gotten better images out of those sensors.

Though it is true that these cameras are starting to encroach into the territory that was once occupied by medium format, the EOS 5DS will likely not come as close. There is such a thing as the "medium format look" and it doesn't come simply from having more pixels. It's a combination of having many large pixels and the colour depth that can be achieved because of it. The 5DS only hits the resolution part of this equation; the acuity bite that comes from being able to use larger pixels and the deeper DR/colour depth isn't completely there, but it's getting close. On paper, it's a better DSLR, but not a genre-breaking one. In other words, if you must stretch reason and your budget to get 50mp at all cost, then doing so with the Canon EOS 5Ds is less of a bad idea than stretching further for a medium format system.

With thanks to Broadway Camera

No comments:

Post a Comment