Points in Focus Photography

Preliminary 6D mark II Dynamic Range Not So Hot

0

When push comes to shove, I have no idea what’s going on in the minds of the management in Canon’s imaging division. A New Zealand photographer, Rob Dickinson, got an early sample 6D mark II from Canon New Zealand, at least as I understand it, and he’s posed some early images from it. Images from which some people in various communities, most notably Fred Miranda[1], have started looking at them to determine the dynamic range of the camera.

On one hand, I do feel it’s important to point out that the images that all this discussion is based on are from a “pre release camera.”

That said, the 6D mark II is supposed to be released on the 27th of this month. That’s less than 3 weeks away. Right now, Canon is building up inventory to ship so that there are more than a camera to have on shelves on release day. The hardware is all final. There can’t and won’t be any tweaking to the sensor or any other hardware between now and the 27th.

In fact, the only thing that Canon could tweak is the firmware, and even that would almost certainly require a post launch user applied update.

In short, my expectation is that the retail cameras will reflect the performance that’s being seen in the sample images from this particularly pre-release sample.

When Canon announced the 6D mark II, I expected them to release a camera that preformed at least on par with the 5D mark IV, if not slightly better.

My reasoning for that presumption was pretty straight forward. Dynamic range is the range between the clipping point (full well capacity) of a pixel, and the noise floor for that pixel (the point where noise dominates signal).

While there are several types of noise that combine to become the noise floor, read noise generally dominates. Further read noise itself is going to be a product of the actual design of the pixel, the read amps, and the ADCs.

One of the biggest improvements that came with the 80D, 1DX mark II, and 5D mark IV, is that Canon has finished their upgraded fabs, and now is designing chips with on-chip ADCs (potentially among other improvements).

On chip ADCs can yield a significant boost in image quality as they allow for a couple of potentially major noise reducing changes.

First, the analog traces that run between the pixel itself and the amps and ADC are necessarily shorter; being contained totally in the sensor. Shorter traces reduce the potential for induced noise from other sources in the camera.

Secondly, and perhaps more importantly, the move to on-chip ADCs allows for many more ADCs to be used to read the sensor in parallel. A general rule of thumb is that the faster the ADC has to operate the noisier it will be.

If you take, for example, the 5DS, with it’s 53 MP total sensor, 5 FPS frame rate, and 16 channel readout we can do some quick estimating on how fast the ADCs might have to operate. At a minimum that’s 16.6 mega samples per second per ADC [2]. In practice though, because some time will be consumed by the exposure, it would have to be faster than that.

On the other hand, on-chip ADC cameras typically have 1 ADC for every row or column. This dramatically reduces the number of samples per second any given ADC has to operate at to meet the performance requirements of the camera.

For example, a camera like the 5D mark IV, with it’s 31.7 MP (total) sensor and 7 FPS drive rate would need to be able to convert 222 mega-samples/second. If the sensor uses row parallel ADCs, any given ADC would only need to manage around 50K samples/second; for column parallel, ~33K samples/second.

For an on-chip ADC camera, the ADCs can be run at much lower frequencies (more of them doing the work in parallel), which means that they can potentially have much lower self noises.

To improve dynamic range, there are basically two options:

  1. You can increase the well capacity (raise the top end limit)
  2. You can decrease the read noise (lower the bottom end limit)

Generally speaking though, significantly smaller looking improvements in read noise will have a much bigger overall effect. Going from 2 e- of read noise to 1.5 e- while still holding a full well capacity of say 10,000 e-, increases the dynamic range from 5000:1 (~12.3 stops) to 6,667:1 (~12.7 stops). However to do the same with well capacity would require increasing the well from 10,000 e- to 15,000 e-, which may very well not be possible due to the size of the pixel itself.

Getting back to the 6D mark II.

Canon said it would be built using the same modern sensor process technology that the 80D, 5D mark IV, and 1DX mark II, used. On top of this it has a lower resolution and frame rate than the 5D mark IV.

The lower resolution means that the pixels would be physically larger, which should mean they can have a higher full well capacity (clipping point). On top of that, the lower frame rate and resolution should mean that the ADCs can be operated at a lower frequency which should give better read noise performance. Finally, using the same sensor architecture should mean that as a baseline the sensor’s read noise should be similar to the 5D mark IV, all things being equal. So equal baseline, plus bigger potential well, plus lower potential read rate and therefore read noise, should equal higher dynamic range.

Unfortunately, that’s not what the early indications are showing.

Early calculations are suggesting that dynamic range may in fact be worse than the original 6D, and considerably worse than the 5D mark IV.

Now again, I haven’t done the calculations for this. However, the two sources seem to have a reasonable track record. Moreover, it’s two people doing calculations — all be it from the same data — coming to similar conclusions. Moreover, their past calculations have jived with tests on release cameras from DXOMark.

So the question is, what the heck is going on and why would Canon do something like this?

Obviously the first point here is that there’s a possibility that the data we have now is wrong. I don’t have much confidence in this, or at least I think it’s a very slim chance, but it is a non-zero probability. That said, I wouldn’t hold my breath on this being the case.

I think the much more likely end is that this early data will be reflective of the camera’s final performance. In which case the situation is really disappointing.

One of the internet theories for this poor performance is that Canon is protecting sales of their 5D mark IV.

To be honest, at least on its face, this does make a little sense. At least I can see the reasoning.

As I said in my first post on the matter, the 6D costs 60% of what a 5D mark IV does, and appeared to deliver more than 60% of the capabilities. If the 6D mark II’s dynamic range is dramatically less than the 5D mark IV, that makes a significant performance incentive for buying the 5D mark IV.

That said, Canon isn’t the only company in the world making full frame interchangeable lens digital cameras. This point is obvious to all of us photographers, and I find it impossible to believe that Canon doesn’t recognize the point too. Nikon is still in business, and certainly have always been Canon’s biggest rival. Never mind Sony appears to be doing everything they can to make major inroads into the camera market. When push comes to shove, both the A7 series and D600 series are expected to be updated this year.

Even if they weren’t, it’s seems reasonable based on past behavior to expect the next 6D upgrade to come along in the early 2020s (maybe 2022), which means the 6D mark II needs to have at least some legs. Admittedly the rate of change in sensor performance has certainly diminished, a camera released now isn’t showing a massive improvement in performance over camera from 3-4 years ago (excepting Canon’s cameras due to the sensor architecture change). In 3-4 years, I can’t say I expect to see major improvements over a camera from now.

That said, based on the preliminary numbers the 6D mark II is starting with a major disadvantage as it stands. In a couple of years, that’s only going to put it further behind. And that just doesn’t seem like a good strategy to me.

Now I do feel compelled to point out that dynamic range and objective image quality testing is not, or at least should not be, the definitive factor when it comes to choosing a camera. Nor will a stop or two less dynamic range compared to a 5D mark IV, D810, or A7R II, render your images worthless. A lot of times, it’s necessary to step back and remember that a lot of the longest lasting and emotionally resonant images were shot on mediocre quality film that resolved less than a modern DSLR does with poorer quality lenses than even the cheapest kit lenses we have now.

I make this point not so much to excuse Canon for what appears to be happening here, but to remind us that the photographer and the subject matter is far more important than the camera.


  1. http://www.fredmiranda.com/forum/topic/1497323/29 This thread starting around this page. ↩︎
  2. ~ 53×106 x 5 FPS / 16 ADCs ↩︎

Comments

Vincent

Hi,
How much better is the read noise of the 5d miv than the original 6d? If at all. I have been unable to find clear information on this!
I’m interested in the 5d miv for astrophotography, which the 6d is often raved over.

Best!

    Jason Franke| admin

    Hi Vincent,

    If you haven’t already found them, you might want to look at the numbers posted on http://www.photonstophotos.net/. He has measurements of the read noise both in electrons and digital numbers.

    That said, I don’t have the practical experience with astrophotography or the 6D to make any kind of reasonable assertions either out of experience or based on someone else’s data.

    Sorry I can’t be more help.

Our cookie and privacy policy. Dismiss