An Update and Comparison of the Sony FE 70-200mm f/2.8 GM OSS

About a month ago, Roger posted his MTF bench results for the Nikon 70-200mm f/2.8E FL ED AF-S VR (TL;DR: it’s optically superb), and down in the comments there were some requests to compare it not only to the Canon 70-200mm f/2.8L IS II in the article, but also to the Sony FE 70-200mm f/2.8 GM OSS. There were MTF charts linked:

These are not very good looking results, especially for a $2600 lens, but Roger wasn’t sure if they were correct results. In his comments after these charts he said:

Let’s kind of keep this here for right now – I haven’t published it because I’m still a little uncertain about the results. Sony has suggested that a change in cover glass thickness might improve the results some. Not dramatically, but some, off-axis. This lens also has to focusing motors and we have to focus it electronically via a camera to test. I’m not absolutely certain that ‘setting it at infinity focus on the camera’ and ‘manually focusing on a an object at infinity’ are absolutely the same. So take these lab results with a grain of salt. On the other hand, they do seem to agree with what we see in real world results.

I try hard not to put out anything until I’m just absolutely certain our results are correct. We’re doing some stuff here that’s pretty cutting edge, honestly. No one does 4 rotation MTFs, for example. I’m pretty certain these are good results, but not absolutely certain. So I’ll post them in this discussion on my own site but I’d rather not see them reproduced all over the internet yet.

Someone suggested we try all three 70-200s on their respective camera bodies and shoot the same detailed scene with each, then share the files. But that would involve three different cameras and wouldn’t really be an apples to apples comparison, would it? Roger suggested he have one of our photo techs shoot all three lenses on a Sony camera with adapters, admitting that adapters add another variable, but there would be some good practical implications.

So I did just that. I took all three lenses, the Canon 70-200mm f/2.8L IS II, the Nikon 70-200mm f/2.8E FL ED AF-S VR, and the Sony FE 70-200mm f/2.8 GM OSS, and I shot our test chart with all of them on the Sony a7R II. For Canon, I used a Metabones T Smart Adapter IV, and for Nikon, I used a Novoflex Nikon to Sony E adapter. I manually focused all three lenses to get the most consistent results, and here’s what I got:

Full Resolution Examples are Available Here

If you view the charts at 100%, you’ll see that they’re consistent with the MTF charts Roger generated for all three lenses. The Sony just isn’t that great, and that’s really disappointing considering the price tag and how long customers have had to wait for that lens to be available. But if you need f/2.8 and working autofocus, it’s really the best option out there for Sony mirrorless cameras.

Author: Joey Miller

I’m Joey. I love cameras, especially old film cameras, and I can’t remember the last day I didn’t take a photo. Digital cameras are great, and they keep me employed, but I also still like processing my own film. I’m stuck somewhere in the middle. I shoot every single day, no matter what.

Posted in Equipment
  • Munchma Quchi

    Affects all cameras.

  • Photographer100

    WRONG, there is no such nonsense as a “perceptual megapixel”
    Lens rentals provided the PROOF above from MULTIPLE lens copies
    DXO mark provides NOTHING……..100% NOTHING but their flapping lips

  • Thinkinginpictures

    DXO has a proper place if you understand their methodology. Have to take it in context. Besides, plenty of independent reviewers have stated the same misgivings over lensrentals findings. I’m sure Roger will follow up at some point.

  • Photographer100

    BS……lens rentals tested multiple copies WITH TEST PICS above
    DXO mark does nothing but flap their lips and has test shots of NOTHING……EVER

  • Photographer100

    thats based on the nonsense filth of DXOmark….BOGUS

  • Photographer100

    !*!*! ~ LAW OF LENS REVIEWS: Whatever DXOmark says…….the OPPOSITE is true *!*!*
    Lens rentals provided TEST SHOTS…………DXOmark NEVER PROVIDES ANYTHING
    Lensrentals tested MULTIPLE copies.

  • Photographer100

    DXOmark is NOT trusted by even a SINGLE professional photog…..theyre worse than awful

  • Photographer100

    AHEM! Lensrentals said the Sony was BAD…..
    Lens rentals provided TEST SHOTS…………DXOmark NEVER PROVIDES ANYTHING
    Lensrentals tested MULTIPLE copies.
    DXOmark is bogus filth
    !*LAW OF LENS REVIEWS: Whatever DXOmark says…….the OPPOSITE is true !*

  • Brandon Dube

    They didn’t test Nikon’s new 70-200. There is also no context for their measurements. What color light is used? What focusing distance? There is no way to predict performance at one focusing distance from information about another. Sony’s 90mm macro is kind of crap at infinity, but a lot better at macro distances. Does that make a test that reports the performance as crap at infinity wrong?

  • EcoR1

    Now the official dxoMark results are out and they indicate problems with Lensrentals test methods. According to dxoMark the Sony 70-200 GM is sharpest 70-200 lens ever tested from any brand.

  • Thinkinginpictures
  • Lotus Eater
  • Do you have any plans to test a later badge of the lens in the future?

  • Brian Smith

    You found 5/6 Sony a7R IIs to have sensor/flange alignment issues? What about other brands (i.e. Canon, Nikon)?

  • Athanasius Kirchner

    Well, that’s a lot of information, thanks for posting it.

  • Carleton Foxx

    Thanks for the reassurance. I’m sure it would make a difference, but I don’t really shoot anything that’s flat, so perhaps it’s not an issue. It does raise the question of how long do camera bodies stay in reasonable alignment. Do they start to sag after two years, three years? Or do bodies stay aligned as long as you don’t mistreat them.

  • Brandon Dube

    I am quite well versed in most aspects of optical design.

    telecentricity is not a design principal, it is a specification. A system is fully telecentric either in image or object space if the chief ray angle is zero One would specify the chief ray angle, for microscope objectives (of all kinds, there is nothing special about confocal microscopy in this regard) it is commonly 0.2 degrees. Telecentricity is desirable in microscopy because it means there is no change in magnification with defocus. Absolute scale is often important in microscopy, so it is important small defocuses do not alter the size of something. For a microscope, the object is usually the aperture stop as well.

    To make a system telecentric, you place the aperture stop at the focus of the elements in front of it, or behind it. If it is at the focus of the elements in front of it you create an object-space telecentric system, the elements behind an image-space telecentric system.

    In a telephoto lens, the rear member has a focal length longer than that of the front member. This can be found through the lagrange invariant; the change in the marginal ray angle from the front member is larger than that of the rear member, so the focal length of the rear member must be longer.

    For e.g. a 100mm lens, the rear member may have a focal length of -75mm. This would require the aperture stop to be three inches in front of the rear member. A telephoto lens by definition is shorter than its focal length, Q.E.D. the aperture stop would have to be in front of the front member as well.

    Reduced ray angles in the corners of the picture is not inherently advantageous. Film, for example, does not care what angle the rays are incident at. A symmetric wide angle has a large chief ray angle, but is easier to correct than an inverted telephoto which will have a lesser chief ray angle.

    Telecentricity somewhat increases track length by restricting aperture stop location, but the real price is paid in the diameter of the elements. If the chief ray angle is zero and the rear element is the diameter of the image you have full vignetting. The rear element must be the size of the exit pupil in addition to the size of the image.

    Generally, an increase in the aperture also increases the diameter. This compounds with the increase in diameter from telecentricity. The short flange distance of E/FE mount is irrelevant to such a discussion, as the rear element can be placed any distance from the sensor greater than the flange distance. These telecentric cinema lenses from Schneider are just as telecentric in front of an E mount as they are a PL mount.

    Because light tunnels into silicon a significant distance before abortion and pixels have finite width, a sensor has what is known as an angle of acceptance, which is the highest angle of incidence which will be properly absorbed by a pixel. You can find some information here:

    Because the capacitors and other complementary circuitry for each pixel are placed behind the photodiode in a BSI sensor, there is less ‘shading’ due to electronics, and the angle of acceptance is larger. This is quite good, as it alleviates constraint on wide-angle lens design. There is an increase in noise for a BSI sensor, as the electronics are more closely packed together, which results in a loss of dynamic range. The natural progression of things is towards all sensors being BSI; I would not say the A7rII “had” to be BSI. It is just Sony’s first camera with a large sensor to utilize the technology.

    Aspherical lenses would never be applied to improve the out-of-focus rendering of a lens. They are used to compensate aberrations an raise the performance of a design, or reduce the number of surfaces needed to attain a given level of performance, reducing size. Sometimes both of these can be obtained simultaneously.

    Aspheres are more costly to fabricate and test than spherical lenses. Often, they are not necessary. Generally, the older and more traditional a a company or designer, the more infrequently they will use aspherics. In optics Sony is young and they are a technology company at heart. It should not be surprising they use a comparatively large number of aspheres compared to e.g. Canon or Nikon.

    The 85GM is quite massive, and is a very poor example of using aspheres to reduce size.

    P.S., you are fond of photozone – you may find some articles on lens design I wrote for them some time ago.


  • Sator Photo

    Somebody doesn’t know the difference between telephoto lens design and telecentric lens design!

    Application of telecentric design principles to photography lens optics is highly unusual. It’s more the sort of thing you’d find in a confocal microscope. Optical engineer, Egami (in his now AWOL blog) explained that making a lens more telecentric does have the advantage of making the light rays more parallel when hitting the corners of the sensor. The disadvantage is that it causes a blow out in lens size. The larger the maximum aperture, the steeper the corner angle of light incidence, and the worse the compensatory size blow out necessarily becomes. Egami explained that this is the reason why designers fight shy of ultra fast lenses for the FE mount, as it tends to disproportionately impact on the negative size blow out of the lens.

    Sony engineer Ken Nakajima, from the a7RII design team, confessed in an interview with Japanese magazine CAPA (July, 2016 issue page 171) that the reason the a7RII sensor had to have a BSI design (despite this resulting in significant loss of dynamic range) was to compensate for the corner shading that the steeper angle of light incidence from an ultra short flange distance causes. Nakajima explained that this is because, unlike film, digital sensors have a three dimensional structure, and this means they cannot register light hitting it at too steep an incident angle.

    The use of BSI sensors and telecentric optical design principles are ways of compensating for the steep angle of corner light incidence on ultra short flange distance mounts such as the FE mount. But these compensatory measures have their own unintended consequences.

    Although Sony’s advertising department made a big song and dance about how their new generation aspherical glass doesn’t degrade bokeh as much, the recent Photozone review of the Sony FE 24-70mm f/2.8 GM also cast graves doubts over the claims to superior bokeh from their new aspherical elements. It is far more likely that their prominent use of aspherical elements in FE lens design has other optical motivations. Namely, in the fact that aspherical elements can be used to make a lens smaller, which in this case may be necessary to partially mitigate the size blow outs inherent to fast FE mount lens designs (which Photozone amusingly called “the toad you have to swallow”).

  • Sator Photo

    Lens Rental testing by Roger Cicala has previously confirmed image degradation from the use of adapters:

    However, I think there is still some uncertainly as to the exact mechanism of this. Micro-tilt is certainly one of them. I also suspect that design mismatch resulting from a DSLR mount lens being placed on a mirrorless mount is another of them. There may also be difference arises from changes to the sensor microlens or in the design of the coverslip glass. More interesting would be to empirically document with MTF plots the exact degree and characteristic of degradation from the use of adapters.

  • KuriousOranj

    This sort of stuff is basically why I’m back to Canon, in spite of the benefits of the EVF and sensor tech. I’m a serious amateur who can’t afford a second FF Sony as back up so 2 trips to the repair centre (i.e. 7 weeks lost from less than 100) in under 2 years for my A7r was not an acceptable performance.

  • Brandon Dube

    A telephoto lens will not attempt to be telecentric; the rear member has net negative power, but not a large amount. The aperture stop would have to be very far in front of the front group to be telecentric. Unless there is a huge airspace in the lens, it is impossible to make a telecentric telephoto.

    There is not much need to try to be telecentric in a telephoto lens, because the angle of view is small the chief ray angle is constrained to be small. Mapping something 10 degrees off axis onto the image certainly isn’t going to mean e.g. a 25 or 35 degree chief ray angle like you may find in a 90-degree FoV wide angle lens.

    The refining process isn’t pioneered by Panasonic, though they may hold that honor in the consumer ring. A lot of pioneering work on super smooth aspheres was done by Itek corp, which is now dissolved with IP owned by a number of companies including Harris and Raytheon, among others. In the 80s Zygo’s precision polishing group did a lot of work there. Later in the 90s, QED technologies created the MRF process, which is among the smoother technologies available today at reasonable prices.

    I disagree with the assertion that most aspheres today (or even 3 years ago when Panasonic announced their new process) are molded; the majority are still ground and polished using a ring tool geometry on any number of commercial CNC platforms whether it’s statis-loh, optipro, opto-tech, schneider, in house, or a different manufacture. There are only 8 or so optical glasses and plastics that can be molded, out of the 150 or so options available. None of them have a high refractive index, and none have very low dispersion. That is quite restrictive in the design process.

    Molding is far cheaper; but it is the developing technology, not the established one.

  • Munchma Quchi

    Get Imatest and make sure that you have the target aligned and a stable test bed. Use a variety of lenses and all processes to avoid affecting the results. If you start to see a pattern of lower scores in one corner or one side you can guess the flange and sensor isn’t aligned. Sure it can happen in any camera. Especially after a drop with a heavy lens.

  • Athanasius Kirchner

    All telephoto zoom designs try to maximize telecentricity. On DSLRs, it’s almost a requirement because of the large flange distance, hence the retrofocal wide angle designs. This ridiculous legend about the short flange distance hurting E-mount needed to die yesterday.
    Sony chose to use asphericals because they’ve invented a refined process for their molding which supposedly eliminates the dreaded “onion rings”. Panasonic pioneered this approach in some Leica-branded designs, and has reportedly worked very well.
    I do agree that these results do not hold up to Sony’s boast of “future-proof” lenses, designed for 100 MP sensors. Either something’s wrong on their assembly line or their engineers are high on the good stuff.

  • Timothy Renzi

    I suggest you forget you ever read that and do not attempt to do so, lest you spend the rest of your life fretting even more about every piece of equipment you buy.

    Just MHO

  • decentrist

    Thank Roger for the testing and stop the whining. The Sony does not perform according to it’s price.

  • David Kilpatrick

    Your tiny concern was what I had in mind. The whole Sony E mount system seems to depend on self compensating setting of the lenses. Personally – I’m very happy with their 70-300mm G OSS, which given its exceptional close focus is one of the best (despite the steep reduction in max aperture as you zoom). It’s as big as I ever want to go! Never tried it on a chart. Real life subjects have been sufficiently good not to need to diagnose any problems.

  • May I remind everyone that on this blog we discuss facts and do not make personal aspersions about other posters please. I know sometimes things are meant to be funny, but it doesn’t always come across that way.

  • I nearly wept in gratitude. I’ve been talking about this for years, but no one wants to hear it.

  • David, see above. We have a body wired into the bench when we test Sony lenses.

  • We’ve really never seen major changes in QC optically. We have seen electronic issues addressed, like bad IS units, inaccurate AF, etc.

Follow on Feedly