A Second, Hands-On, Look at the Canon EOS 5DS

A few months ago I wrote a brief preview of the new Canon EOS 5DS and 5DS R that covered the basic camera specifications and the similarities to (and differences from) the EOS 5D Mark III. Recently I had the chance to shoot for a couple days with a pre-production sample of the EOS 5DS. It’s important to note that this was a pre-production sample and it’s possible there could be some small changes made before the production version is shipped, though at this stage the changes are likely to be in software, not hardware, since the camera is due to ship next month (June 2015). While I didn’t have time to fully evaluate all aspects of the EOS 5DS, I did look at what most people will regard as the most important ones, namely resolution, noise, and dynamic range.

Resolution

The 5DS has the highest pixel count of any full frame DSLR to date, 50.6 MP. That’s 2.27x as many as the EOS 5D Mark III (22.3 MP) and 2.5x as many as the EOS 6D and 70D (20.2 MP). That should result in about a 50% increase in resolution over the EOS 5D Mark III.

The question then is whether the theoretical increase in resolution can be seen in practice and if so, is it only seen when using the best lenses at their optimum aperture with everything mounted on a sturdy tripod to minimize vibration? I set out to answer this by comparing shots taken with the EOS 5DS to those taken with the EOS 6D (also full frame but 20.2 MP) and the EOS 70D (APS-C, but at 20.2MP it has about the same pixel pitch as the EOS 5DS). For a lens I used the EF 85/1.2L stopped down to f/4—an aperture at which it is very sharp. At the other end of the lens scale I used an old EF 22-55/4-5.6 plastic lens mount kit lens that was sold to go with the EOS IX, an APS film camera.

Here’s the first set of comparison shots, taken using the EF 85/1.2L stopped down to f/4:


Note: 5DS sample image shot with a pre-production beta camera. Image quality might not represent final quality.

On the left is a 100% crop from the EOS 5DS. In the center is a 100% crop from the EOS 70D and on the right is an upscaled image from the EOS 6D, scaled to match the other two images.

Well, it’s pretty clear that the EOS 5DS shows much better resolution than the EOS 6D, as would be expected, with a good lens. It’s also clear that the EOS 5DS and EOS 70D show similar resolution, though of course the EOS 70D is APS-C and so has a lot less coverage than the EOS 5DS. It follows from this that you could take an EOS 5DS shot and crop it to APS-C dimensions if you wanted the “1.6x focal length multiplier” effect that you get with current APS-C cameras, without any significant loss in image quality. With the 5DS you can afford to use a shorter lens and crop the image.

So much for 100% crops viewed on a monitor, but what about prints? To see what the difference would be, I made prints of small sections of the 5DS and 6D images blown up to 20″ × 30″, which is a fairly large print for most people. Looking at the prints from the distance of the print diagonal (36"), I had trouble seeing any significant difference in sharpness. However, when viewing from closer distance (12"), the print from the 5DS looked sharper. So if you ask how large a print you can make from a 5DS or a 6D before you can see a difference, you have to also say how closely the print will be viewed (and how good the eyesight of the viewer is at that distance). The larger the print and the closer the viewing distance, the more obvious the resolution advantage of the 5DS becomes.

Of course that’s all well and good if you have a $2000 lens shot at optimum aperture and look at the center of the image. What about a cheap lens and the edge of the image? Will the 5DS actually give you any benefit over a camera like the EOS 6D or EOS 5D Mark III? That’s answered in the next set of crops, which were taken from the edge of an image shot with the EF 22-55 wide open at 55mm (f/5.6).


Note: 5DS sample image shot with a pre-production beta camera. Image quality might not represent final quality.

The upper image is a 100% crop from the EOS 5DS, while the lower image is a 100% crop from the EOS 6D, upscaled to match the 5DS image. As you can see, even with a low cost plastic mount kit lens from the 1990s shot wide open and sampled near the edge of the frame, the 5DS image clearly shows more detail.

The conclusion here is that the 5DS will show higher resolution no matter what lens (within reason) you use it with. You might gain more from a high quality lens, but you still gain even with a consumer grade kit lens. The 5DS image is clearly better every time.

Noise & Dynamic Range

Now we get to dynamic range. I couldn’t do any real quantitative measurements, but I did take a series of images at ISO 100, with increasing under exposure to -6 stops and processed them using RAWTherapee to digitally “push” them back to the correct exposure (Canon’s DPP only allows for a 3 stop correction). Note that a -6 stops underexposure at ISO 100 is the same as using the correct exposure for ISO 6400. Here’s the result:


Note: 5DS sample image shot with a pre-production beta camera. Image quality might not represent final quality.

These are 100% crops from direct RAW conversions with no noise reduction applied. At -3 stops things don’t look too bad, but by -5 stops there’s quite a lot of noise and by -6 stops you’ve passed the normal limit of acceptability. Even with some serious noise reduction the image at -6 stops isn’t going to look good. This is pretty much in line with what I’d expect from Canon’s DSLRs. You’d see much the same thing from the EOS 6D or the EOS 7D Mark II, though it’s perhaps a little better than you might see from the EOS 5D Mark III. It’s been speculated (and probably established by experiment) that this noise appears to come from the off chip Analog to Digital (A/D) converters. Nikon DSLRs, using Sony sensors with on chip A/D conversion, show better performance in this respect, with even -5 to -6 stop pushes of the RAW images giving acceptable results.

What you can take from this is that the new 50.6 MP sensor in the EOS 5DS is almost certainly using the same basic architecture as the sensors in Canon’s other DSLRs.

Comparing the -6 stops underexposure at ISO 100 to the same exposure at ISO 6400 shows the following:


Note: 5DS sample image shot with a pre-production beta camera. Image quality might not represent final quality.

[Note: The image scale is different because the two shots were taken at different distances from the subject.]

As you can see, it’s WAY better to shoot at ISO 6400 than to shoot at ISO 100 and try to “push” the image by 6 stops in software. The right side image shows the noise performance of the EOS 5DS at ISO 6400 with no noise reduction applied. This is pretty good performance, probably on par with the EOS 7D Mark II but not as good as the EOS 5D Mark III.

One odd thing about the ISO settings on the EOS 5DS is they stop at ISO 6400 (12800 with expansion enabled). Clearly this isn’t due to poor noise performance. The 7D Mark II has ISO settings up to 51200. Why Canon decided to cut off the access to higher ISO settings in the EOS 5DS is something of a mystery. I can only guess it was some sort of marketing-driven decision. While the audience for this camera (landscape photographers, studio photographers, product photographers) might rarely, if ever, want very high ISO capability, it seems a shame to deny it to those who could use it. If it’s a software issue, it could possibly be addressed by a firmware update—though of course that doesn’t mean it will be!

I haven’t yet had a chance to shoot with the 5DS R model, which has a different filter in front of the sensor that essentially removes the effect of the standard low pass filter. This should provide even higher resolution, though at the expense of potential moiré problems on areas with fine patterns.

I didn’t have enough time with the camera to evaluate things like exposure and autofocus or try out all of the bells and whistles of the Canon EOS 5DS, though I can say that while I had it, it performed flawlessly. It was responsive, AF was fast and accurate, and the image quality was remarkable.

Sign in or Sign up to post response

    • Hmmm, at $3699.00 sounds like a steal.

    • To post a reply Sign In
    • Good article but I fail to see the purpose of enlarging the 6D image to match the other 2 in the first set of 3 images. Completely pointless.

    • To post a reply Sign In
    • No, it's not pointless. If you understand the math, it shows how similarly sized prints would look if viewed from close up. If you just compare 100% crops you'd be comparing, for example, the equivalent of a 16x24 print from a 5Ds with a 10x15 print from a 6D.

      What may be confusing is that that's actually what you are doing when you compare the straight, uncompensated, 100% crops from the 5Ds and the 70D. They look like they have the same resolution, but the 5Ds print would be larger by a factor of 1.6. The comparison just shows the two sensors have similar native resolution.

    • To post a reply Sign In
    • Hi Bob,

       

      I understand the maths and I understand that both the 5Ds and 70D have the same pixel density. I just think it was pointless to show a 100% crop from a 70D next to an enlarged crop from a 6D. Both cameras have the same resolution so surely showing 100% crops from each or enlarged crops from each would make more sense. 

       

       

    • To post a reply Sign In
    • Jamie, I think Bob was trying to answer two different questions: (1) is the per-pixel sharpness of the 5Ds similar to what we've seen from the APS-C cameras with identical pixel density, and (2) is the 5Ds actually able to recover more detail from a scene than the lower resolution full frame cameras. Comparing it to a non-enlarged 70D image and an enlarged 6D image would be the correct way to answer those questions.

      So, nice preview. And I'm amazed how well that old 22-55 performed.

    • To post a reply Sign In
    • pointless to show a 100% crop from a 70D next to an enlarged crop from a 6D. Both cameras have the same resolution

      The 6D and 70D don't have the same resolution. They have essentially the same number of pixels, but the pixel density of the 70D is higher than that of the 6D. In fact it's about the sames as the 5Ds.

       To compare the 70D with the 5Ds as far as what equal size and equal coverage prints look like from the two cameras, I'd have either had to use a shorter focal length lens on the 70D (53mm, thus a different lens with different sharpness)  or move to a distance 1.6x as far away (not practical at that locations), then upsize the image to match the 5Ds.  

      However I wasn't trying to compare ultimate print quality from the 70F and 5Ds. I was looking at how the native resolution of the 5Ds sensor compares with that of the 70D sensor. 

      For the 6D I could do the "equal print size with equal coverage" comparison. So I did!

    • To post a reply Sign In
    • I understand you were demonstrating the native resolution of each sensor but my argument is that it is fairly pointless to show it. An 8MP iPhone has far greater pixel density than any DSLR (including the 5Ds) but who would want to see an upscaled crop from an 8MP DSLR alongside a 100% crop from an iPhone? It provides nothing of value when judging which device gives more detailed images. That was my point.
    • To post a reply Sign In
    • You guys arguing about the same thing, it is just demonstrate that Canon 85/1.2 in optimal resolution out resolving 20mp 6D sensor.
      But in real life frame size is the same 24x36, price point getting closer to medium format cameras which is different ballpark completely, 50 mp medium format back will beat 24x36 at any given day.

    • To post a reply Sign In
    • Crop sensor (i.e. 1.3x relative to 645 film) medium format cameras which have a slightly larger sensor than 24x36mm are "only" about 2-3x the cost of the 5Ds (R), however they have the baggage of the lenses having been designed to cover full frame medium format (i.e. at least 6x4.5cm or 6x6cm), so wide angle availability and cost is a problem. Full frame medium format (645) costs about 10x the price of the 5Ds, just for one camera body with sensor. If you can afford full frame medium format digital then wide angle is not a problem, but we are then talking about different orders of magnitude cost relative to the 5Ds (R). And if you need to photograph moving subjects and focus track them with shallow depth of field,  the 35mm full frame cameras probably do the best job at that (and if the MF DSLR system that you're considering supports AF, the sensor or sensors are in the center of the frame - remember the early AF cameras?). Action photography is actually a good application for high resolution 35mm digital SLRs since one can keep the framing slightly loose and not lose as much quality when finalizing the composition in post. You get extra range for your subject to move in, before framing or image quality becomes a problem. Of course 5fps is slower than 10-12 fps, but it is still fast enough for most subjects with an experienced operator. One practical issue is the amount of data generated when photographing action with a 50MP camera.

    • To post a reply Sign In
    • Ilkka, Pentax 645D is "only" $4289 in B&H and sport application of 50 mp probably will be limited to "Sports Illustraded".

    • To post a reply Sign In
    • Yay!! GO CANON!! This is what we Canon users were expecting for a long time!!

       

      As for the comments, yes it's obvious (or it should be) that a medium format (bigger) sensor with medium format lenses will give more resolution and sharpness even if the sensor technology and megapixel count is identical, but so what? We want to use our Canon EF lenses and extract the maximum out of them, and the 5DS delivers!!

       

       

    • To post a reply Sign In
    • Spoiler alert: non-pixel-peepers can quit reading this comment right now. For the more technically inclined, please read on.

      It would have been interesting to see the comparison to a Canon 5D. I expect the 5DS would demonstrate significantly higher resolution. If so, it would put to rest claims that existing (i.e. prior to the 5DS) digital cameras were able to record all of the resolution available from currently available lenses.

      Off the top of my head I suspect that the 5DS probably gets closer to wringing out all of the available resolution, but is not all of the way there yet.

      Let's assume, for example, that a lens is able to resolve about 100 line pairs per mm. To realize that degree of resolution in the image would require a minimum of 200 sensors per mm. That would work out to about 35 megapixels in a 24mmx36mm image. This would make one think that a 50 megapixel camera would get us there, and further improvements would be pointless.

      But wait! There's more to the "picture", if you don't mind a play on words. A "50 megapixel" Bayer sensor only has 25 green megapixels, which means that for the green parts of the image we aren't quite at a lens-limited resolution level of 100 line pairs per mm. In fact, the limit for 25 megapixels is about 86 line pairs per mm, which is very good, but probably somewhat under the best lenses can deliver at optimal aperture.

      It's worse for the red an blue sensors, since they are only 12.5 megapixels each for those colors, meaning that they would top out at about 60 line pairs per mm, not bad, but probably somewhat short of what the best lenses can deliver.

      This means that a camera with a "50 megapixel" Bayer sensor is going to give really high resolution images, but still probably a little short of what the best lenses can deliver to the focal plane.

    • To post a reply Sign In
    • Let's assume, for example, that a lens is able to resolve about 100 line pairs per mm. To realize that degree of resolution in the image would require a minimum of 200 sensors per mm. 

      Well, that's the Nyquist theorem and it's about reconstructing sine waves from data points. If you want to reconstruct square waves (which have a large 3rd harmonic content), you'd need 3x that, plus it's an absolute limit, so you need more than the minimum in practice. Then you have to consider MTF (modulation transfer function) which is a measure of how contrast drops as the lens resolves finer detail.

      Put it all together and it's unlikely you'll get near the limit of sensor resolution anytime soon.

      The best lenses in the center of the image field are probably capable of resolving over 400 lp/mm in the aerial image at f4 (sine waves). To resolve that you probably need to be at at least 1000 pixels/mm, which puts a 24x36mm sensor at 864MP.   Some of the small Sony sensors have a 1.12 micron pixel size. A 24x36mm sensor with such size pixels would have about 688MP.

      At that point you really would be sensor limited!

    • To post a reply Sign In
    • Good points Bob, though I am not sure if the best easily available camera lenses really would achieve 400 lp/mm, but if they do then a lens that is even half as good would still far exceed the capabilities of today's digital sensors.

    • To post a reply Sign In
    • I wouldn't be surprised to see quite a few good lenses hitting close to 400 lp/mm in the center of the image (400 lp/mm is diffraction limited at f4).  With a properly aligned and centered lens, the only aberration affecting image sharpness in the center of the field is spherical aberration, and that's not difficult to eliminate in a fairly slow lens (or a fast lens stopped down).  As you move away from the center you get the effects of field curvature, coma, astigmatism and transverse chromatic aberration coming in, all of which will lower resolution, but none of them affect the center.

      This is academic of course, but I suspect that even in practical application, with a lot of real world lenses a 100MP full frame sensor would show significantly more detail than a 50MP full frame sensor when shooting static subjects under controlled conditions. I think someone, if they wanted to, could probably build a 100MP sensor full frame camera with good performance using current sensor technology, storage and processors. In fact I'd be surprised if someone wasn't actively working on it right now.

      I don't need a 100MP camera, but I'm sure someone does (or if they don't need it, they at least want it!).

    • To post a reply Sign In
    • I'm glad to see something to compete with the medium format cameras without shelling out $25K and more.
      I have the 5DIII and until I start putting posters on the side of semis, I think that will fare well for me. Frankly, if I need something sharper, I use a tripod.

      h

    • To post a reply Sign In

Sign in or Sign up to post response