New Items

Canon 5DS and 5DS R Initial Resolution Tests

Published June 15, 2015

Like everybody else, we’re pretty excited to get our hands on Canon’s new 5DS and 5DS R. There are already a lot of hands-on articles about the cameras that probably have told you more than you need to know to make your purchase decision. Of course, for most of the Canon shooters who read this blog, the purchase decision was just which place you want to buy it from.

For me, I want some lab data to see just how much of a difference those megapixels make. More particularly, I want to see how much of a difference they make when shot through a reasonably good lens, an excellent lens, and an adequate lens. Some people want to simplify things too much and claim certain lenses are ‘good enough’ for the new cameras and others aren’t. It’s not that simple.

So we begged and threw temper tantrums until Drew agreed to let us have a couple of the new cameras for a couple of days testing in our Imatest lab. That was enough time for us to get a quick overview using several different sample lenses, but it will be months before we have a good database of which lenses are most capable on the new cameras.

 

Yes, I had to. A cheap cell-phone pick of the highest resolving SLR cameras made just seemed so wrong that it had to be done.

 

There are several significant differences between the “S” and “R” models and the standard Canon 5D Mark III that these new cameras are not replacing. Lots of real reviews will cover this in detail. Both the new cameras have 50.2 megapixel sensors, compared to the 5D Mk III’s 22.1 megapixels. The 5DS R version has self-canceling low pass filters; effectively eliminating the low-pass filter blur. That should give the 5DS R version a bit higher resolution than the 5DS.

There is one point I want to make early and often, though. Determining how much difference the sensor will make when it is one piece of an imaging system is not a simple thing. Some experts will throw out a calculated number, saying you can see ‘this much’ resolution difference when moving from a 5D III to a 5DS or 5DS R. That number will usually be accurate if you are considering just the center resolution. When using a lens with a long exit-pupil distance, like many telephoto lenses, the calculated number may even be accurate off-axis. On the other hand, some lenses will not show much resolution difference in the edges and corners of the image.

Over time we’ll find which lenses take best advantage of the new, high-resolution sensors, and which don’t. Almost every lens is going to benefit to some degree, of course. The idea that certain lenses are “50 megapixel ready” is kind of silly. The 18-55mm kit lens will still have more resolution on a 5DS than it will on a 5D III.

Testing Methods

We used an ultra-high resolution backlit film chart for the telephoto Imatest images, trying to make certain the chart used didn’t limit the resolution of the final images. Given the very high resolutions these sensors can produce with excellent lenses, I had some concern that a printed paper chart could become the rate limiting factor when testing resolution.  (I expect some test labs that use inkjet printed resolution charts are going to find this to be a problem. Linotype prints probably not so much.)

 

Telephoto Test: The Canon 300 f/2.8 IS II

The Canon 300mm f/2.8 IS II is a really sharp telephoto lens that is still small enough to work with easily in the lab. We shot all 3 cameras through the same copy of the lens. Mirror lock up, 10 second timer, focus bracketing, and all the routine techniques used in all our other Imatest articles were used. The table below shows the resolution differences between the original 5D Mk III and the two newer cameras when shot through the same copy of the lens.

  Center Average Corner Avg.
5D Mk III1050920895
5DS 137511901070
5DS R154513001175

Resolution increased about 30% when moving from the 5D III to the 5DS in the center and for the average of the entire surface of the lens, which was about what we had expected. Resolution in the absolute corners increased about 20%. This isn’t shocking since we would expect the corners of the image to be more limited by the lens (although I had hoped such an excellent telephoto lens might do a bit better).

I was a bit surprised at the difference between the 5DS and the 5DS R in this test. The 5DS R version provided about a 10% increase in resolution over the 5DS.

 Standard Range Test: The Zeiss 85mm f/1.4 Otus

We were wanting to make this test about the cameras more than the lenses so we chose one of the sharpest standard range lenses we could think of, the Zeiss 85mm f/1.4 Otus. We repeated the test the same way, shooting at f/2.8 again.

  Center Average Corner Avg.
5D Mk III13001015835
5DS15701145970
5DS R16601240990

With the Zeiss 85mm Otus things are just a bit different. The improvement between the 5D III and the 5DS is about 20% in the center, slightly less off-axis. The 5DS R version adds about 5% more resolution compared to the 5DS. I was a bit surprised at the result and repeated it several times to confirm the numbers and it was so. The reason I was surprised is the Otus is the highest resolving lens, and my first thought was it therefore would benefit the most from the higher resolution cameras.

Wide-Angle Test: The Zeiss 21mm f/2.8

For this test we moved back to our largest printed paper resolution chart and are shooting even then at a fairly close distance (about 9 feet). The Zeiss 21mm f/2.8, old as the design is, is still on of the sharpest wide-angle lenses in the edges and corners.

  Center Average Corner Avg.
5D Mk III1060800
540
5DS13551025680
5DS R14951140690

Again, we see the resolution is greatly improved, with the 5DS MTF50 numbers about 30% higher than the 5D Mk III, even in the corners. Throughout most of the image, the 5DS R MTF50 numbers are about 10% higher than the 5DS. However, in the absolute corners there is only a small difference between the 5DS and 5DS R cameras. From other things we know about this lens, I would assume this is because the corner resolution with this lens is limited by astigmatism.

I want to clarify this just a bit because depending upon how a tester is presenting data, you may see them present this corner data differently. Imatest doesn’t measure astigmatism exactly, but does measure slanted lines that are nearly vertical and nearly horizontal. When I present corner data I average 8 numbers: the horizontal and vertical reading in each corner. Someone else may present only the best corner number, or the median corner number, or just vertical or horizontal numbers. This might have a corner number better or worse than what we are presenting.

The Not So Great Lens Test: Canon 50mm f/1.4

I wanted to add this to make a point. I’m beginning to read people discussing “50-megapixel capable lenses” and I’m afraid it creates some confusion. It’s true, these high-resolution sensors are going to magnify the difference between lenses. But that doesn’t mean they don’t benefit every lens. With this in mind I went and found a battered old copy of the Canon 50mm f/1.4 lens. This is a decent, although few would claim fabulous, lens. This particular copy was also very slightly decentered. It wasn’t even the best of the 50mm f/1.4. I test all three cameras with this lens just like I did the others (except this test was done at f/1.4 — I want the lens to be just as weak as it could be).

  Center Average Corner Avg.
5D Mk III660475330
5DS790565340
5DS R800570350

Even with a weak copy of an average lens, there’s no question the high resolution sensors improve the resolution, at least in the center. The improvement isn’t as great as it is with the excellent lenses we tested above. The 5DS had about 20% better resolution in the center and as an average compared to the 5D III. Corner resolution, though, changed very little. The difference between the 5DS and 5DS R was also negligible with this lens. But there’s no question even a weak lens shows significantly improved resolution on the new cameras.

Summary

Just as it was the Nikon high resolution cameras were released, it will take months of testing and photographer experience to determine which lenses do exactly what with the new cameras. But it’s not a critical thing to know at first. Every lens will have better resolution when shot on the new cameras, at least in the center of the image. How much of a difference will range from noticeably better, to “wow” depending upon the lens, type of photography, and subject matter.
There are many other things that are going to be as, or more, important to the photographer than absolute resolution. But absolute resolution is always a good thing. And these cameras certainly deliver phenomenal resolution.

 

Roger Cicala

Lensrentals.com

June, 2015

Author: Roger Cicala

I’m Roger and I am the founder of Lensrentals.com. Hailed as one of the optic nerds here, I enjoy shooting collimated light through 30X microscope objectives in my spare time. When I do take real pictures I like using something different: a Medium format, or Pentax K1, or a Sony RX1R.

Posted in New Items
  • mike brassey

    My very jaded take on the new cameras
    Conan Photographic Corp camera body dept meeting in the recent past:
    “Money”grutnted the ageing president to a hushed and reverent audience not used to meeting the esteemed man. “While it is true that the Conan corp international group has had so much success in so many other fields that we have perhaps overlooked your small department i thought it wise to conduct some extensive surveys” piped up the newish enthusiastic CEO.”I sent a couple of people to hang around the street corner near that Nokin place and do you know what ,they saw a lot of familiar faces popping in to see their latest cameras”.”Money” repeated the ageing president conscious(but only barely) of the approaching deadline for the next huge payment to the aptly named “Immortality Gardens”built in his honour.He acutely remembered his cousin Monte Olympus’s memorial gardens being reduced to couple of old pot plants and he certainly wasn’t going to allow that to happen to him. The lanky R and D chief slowly rose”I think i have a solution .Why don’t we just glue a couple of 7d sensors together,a bit of extra computing power and we could have er let me see……50.6 megapixels”.The ever so slightly slimey marketing boss shrieked “OMG 50 megapixels “ at the thought of all the marketing opportunities and promptly wet himself.The down to earth accountant finally took the stage .” well but we have all those old mk3 bodies that nobodys going want anymore and the new body will never be designed if you lot in R and D don’t stop playing computer games and do some work , so lets pop it all in them, problem solved”.Charlie ,the junior in R and D couldn’t contain himself any longer and needing at least to say something ,jumped up and blurted out “ we could like have 2 models like,sort of like one with and sort of like one without”.The marketing boss looked as if he needed hospital treatment as he salivated over the idea of selling two models maybe even one of each to the same person.The R and D chiefs head slowly crumpled into his hands while computing the potential problems of aa filter removal,focus stack thickness and a myriad of other thoughts.
    HAS CHARLIE THE JUNIOR BITTEN OFF TOO MUCH?
    IS THE R and D CHIEFS WIFE GOING TO SEE HIM ANYTIME SOON?
    STAY TUNED FOR A THRILLING FINAL EPISODE

  • So AA-less version increases measurable MTF50. But so does sharpening. In D800 vs. D800e tests I have seen some claims that pictures from D800 with mild sharpening were very similar to D800e (for example here: http://blog.falklumo.com/2012/05/d800-aa-filter.html ).
    It might be interesting to apply some amount of USM (unsharp mask) and measure MTF50 again. How would 5DS + sharpening stand against plain 5DS R?
    Of course if you have high ISO, sharpening will also affect noise, but for low ISO it might not be problem.

    Both AA-less and USM will cause some artifacts depending on scene, but with USM you can choose amount of artifacts after picture is taken.

  • Roger Cicala

    Samuel, yes we are. We try to lead horses to water, but that’s about it 🙂
    In this case, there’s been a lot of disinformation put out be a couple of resources who want to dumb everything down to a ‘yes’ ‘no’ answer and a lot of people like simplicity, not matter how incorrect it is.

  • You must be getting millions of “that can’t be” comments about relatively soft lenses getting sharper thanks to a higher Mpix count. “Maybe it shows in a chart but you won’t see it on an actual photograph” and similar. There’s a simple way to address them, wink wink.

  • Wombat

    I think Canon made a big mistake emulating Nikon’s D800 with its AA / half-hearted-AA-less options. They would have done better to learn from it, and just build a real AA-less design like the D810.

    Maybe Canon has to learn the lesson as Nikon did 🙁

  • mcomfort

    I’d also love to se the results of the Canon 50 shot at f/2.8.

    Also I’d like to see it stacked up against a stellar 50mm, like a Zeiss planar or Sigma art 50, shot at the same apertures – to see for example if a Zeiss 50 on a 5DIII tops that Canon 50 on the 5DsR.

  • Mark Lagrange

    I will be curious to hear about the performance tests at high ISOs, since “surprisingly” the new 5DS and 5DS R are both marketed with with top non-extended ISOs that are far lower than the 5D Mark III. I am wondering about the impact in night time (star) landscape work.

  • Mark Turner

    At least for the Zeiss 21mm f/2.8 lens in the center, these numbers scale closely to what you had published in the lens recommendations for the Nikon d800, very close to what the sqrt(50.2/36) would predict vs the 5ds. The Nikon does better than what that predicts in the average number, which could reflect the stack design in the corners.

  • Brandon

    Frans,

    We rarely use Imatest these days – mostly for when we want to measure system performance. The majority of our testing is done on our ImageMaster and OLAF, which are both faster and much more accurate than using Imatest or other camera based test methods.

    James,

    We did not refocus for the corner numbers – it is best to minimize the number of confounding variables. The blur from field curvature is of constant physical size, so it will scale perfectly as the sensor becomes more pixel dense.

    Regards,
    Brandon

  • Quick question:

    Did you refocus for the corners test? I just want to be sure that you’re not accidentally measuring the field curvature.

    -James

  • @Brandon,

    Fair enough. I mostly mentioned those figures as the physical resolution limits, as opposed to the rule-of-thumb of half-of-vertical-number-of-lines estimate.

    Speaking of polychromatic MTF: it is possible to measure MTF of only one Bayer channel, which also helps to avoid demosaicing and chromatic aberration, but then you are performing measurements on images that are quite unlike how the camera is going to be used in the real world. For most practical uses, the “normal” Imatest results are more relevant.

    But if you want more accurate results than what Imatest can offer, please feel free to contact me.

  • Brandon

    Frans,

    These lenses are not diffraction limited, though the 300 is quite close across the frame and the 21 is very close in the center, which does ‘harm’ the result, though I do not see much room for improvement over the 300 – perhaps at a further focusing distance.

    Other than the reality of the sensor’s MTF being less than perfect, losses from a less-than-100% fill factor, noninfinite chart resolution, focusing errors, vibrations/movements, real polychromatic MTF, and a host of other factors are at play.

  • Assuming a photosite pitch of 4.14 micron, green light at 550 nm, a 100% fill factor, and a purely diffraction limited lens, quick-and-dirty theoretical maximum resolutions for the 5DS R would be:
    f/2.8: about 2835 lp/ph
    f/4: about 2552 lp/ph

    But just as with the D800e, we will not see those values in practice. Especially not with Imatest, which tends to underestimate MTF50 a little (unless you turn on the LSF correction)

  • KimH

    Hi Wally,

    thanks for validating my math :o) i used 56MP not 50.6 – everything else was good…

  • Brandon

    Anton,

    I do not know the sensor stack thickness of either the 5Ds(r) or the D810, and I do not know the thickness utilized in the design of the otus lenses. At f/2.8 they perform at a very high level – high enough for the sensor stack to dominate the performance. The sensors are also not competing in isolation, since we cannot guarantee equal raw converter performance (this makes a large difference – see Fuji raw conversion and detail in areas of green for example) and do not know how the lenses are optimized. There was a big enough disparity between the expected result on the Canon and Nikon systems that it through big red flags, so we did not publish that data.

    Regards,
    Brandon

  • Andy

    Can you re-test the 5D III / 300 2.8 II combo by first applying tape over the lens contacts to prevent the camera from identifying the lens and possibly affecting the image in a way that is skewing your test results?

    I don’t see how it can score within a few percent of the 85 on a high resolution sensor, then fall substantially short on a low resolution sensor. If anything the scores should get closer as the sensor resolution drops.

    Please tape the contacts and see if that resolves the discrepency.

  • Wayne S

    Thanks Roger for all your fine interesting work here. I wonder how much the higher frewuency MTFs differ between the S and the R for the Otus?
    Look forward to your teardown and can’t wait to get my copy of the R and try the 500/4 on it and Otus 55 and some Leica R lenses. Thanks again Roger!

  • Wally

    KimH said:
    Great and ineresting – somewhat surprising that the R makes this much difference.
    I fumbled around with some math – would it be correct to say that the theoretical maximum LP/IH is 3050 in the 5DS?

    No the sensor has 5792 vertical pixels, so the best possible resolution is 5792/2 or 2896 lines pairs per picture height. That assumes of course that the test pattern lines up perfectly with the pixels and that you get at least 50% contrast under those conditions since the measurement is a MTF50 number. In reality you will not do that good. Roger’s best number was 1660, which is probably not too far below the best possible.

  • I wasn’t surprised to see improvement from the 50 1.4. Would have been more interesting to see how much it improved at a more favourable aperture like f8 where the lens would be more sensor limited than suffering from imperfections.

    The lack of difference in corner resolution is also not surprising as in the weakest areas of lens performance there would be no data that a higher resolution sensor could improve.

    I’d expect to see the biggest improvement in performance of cheaper lenses in their sweet spots.

    I own a number of fab lenses such as the TSE-17 and 24-II and the Sigma 50 1.4 ART. I’d expect to see less improvement with a lens like the Sigma 50 ART than the Canon 1.4. I have the Sigma 50 ART in Nikon fit and use it on the D800E. On that camera I think its resolving around 30-32Mp (centre stopped down to about f5.6). So I’d be surprised if it improved much on a larger sensor.

    This makes a lot of sense. If the Sigma 50 ART resolves about 30Mp then clearly a camera of up to 30Mp would show an improvement of up to 30% ish over a 22Mp camera like the 5DII.

    So to me this just proves that the new Canon 5DS will not achieve its full performance with current SLR lens technology and that 30-40Mp is the range of feasible resolution. For this reason the MFDB’s will continue to offer a superior resolution and the forthcoming 42Mp Sony chip will probably have a similar effective resolution to the Canon, but with better noise performance.

    I’d be most interested to see how well the TS-E lenses perform with this camera, because this is one of the places where Canon has a unique selling point.

  • Steve Chernicky

    Thank you
    This was a very interesting comparison.
    I was disappointed however that the EF50 F-1.4 lens was not also shot at F-2.8. (Even the Otus was shot at F-2.8)
    This would have given some insight as to how other “average” lens could be expected to perform. (This lens, as is common with others, dramatically improves when stopped down a couple of stops)

  • Flemming

    For the worst lens example you could have an old power zoom. They make current kit lenses look like Zeiss Oti.

  • Brandon

    Hi Mike,

    Can’t answer your first question, we just tested resolution and then the cameras went back out into the rental fleet.

    Regarding canceling the blur filter – the camera’s sensor stack needs to have the same optical path length regardless of whether it includes an optical low pass filter or not.

    The way the OLPF filters work basically involves splitting the light into two very close together beams in either X or Y, then doing it again later in the other direction.

    The “AALess” design simply splits in X or Y, then instead of splitting the opposite direction, the same split that is done first is done, but backwards, to recombine the rays. This keeps the OPL the same, so the sensor stacks “appear” the same when combined with lenses.

    Regards,
    Brandon

  • Brandon,

    Sensor stack thickness is optimized better on the Nikon than the Canon?

    I’m guessing this is the inference as you were expecting better head to head numbers from the Canon.

  • Other testers have pinned the Otus 55 and 85 to have peak resolution for whole f-stops at f 4.0.

    Why 2.8 ?

  • Robert DeCandido

    Thanks! Quick question – how well and quickly do the new cameras focus with telephoto lenses? In other words, could one use them in lieu of a 5D3 for wildlife and/or birds in flight?

  • Chris

    Great news. I wish you would’ve tested a common modern Canon zoom like the 16-35 F4L (which in theory should smoke that Zeiss 21mm) hahah… or a 24-70 2.8 II. Those are lenses everyone will be using. Also I bet if you stop that 50 1.4 down to 5.6 or F8 (or even F4) it will give those Zeiss lenses a run for the money. Stopped down on full frame, the Canon 50 is razor sharp in the corners. Check out Digital Picture’s charts.

  • Greg Dunn

    Thickness of the sensor stack is very important – in fact, Roger did a blog on it recently. http://wordpress.lensrentals.com/2014/06/sensor-stack-thickness-when-does-it-matter

  • Mike

    Roger,
    So (1) why would a person want the 5DS and not the 5DS R and (B) wouldn’t it be better to eliminate the blur filter rather than add an additional filter to cancel it. I heard something about keeping the sensor stacks the same thickness but that seems a bit lame to me.
    Unrelated to the resolution topic, but I thought that EF-S lenses protruded further into the camera and would interfere with the mirror of a FF body. Are you saying that I could adapt my EF-S 17-55 f2.8 to fit my 6D?

    Thanks for everything you do for the photographic community!
    Mike

  • I thank you and your staff for the enlightenment on the 5Ds and 5DsR

  • Roger Cicala

    Chris, et al. Sure it will if you remove the pin (obviously won’t cover the corners). But the point I was trying to make is the crappiest lens will still benefit.

Follow on Feedly