Nikkor 17-55mm f/2.8
It is relatively expensive, and most people buying it rightly has high expectations for this Nikon top-of-the-line pro-grade fixed f/2.8 standard DX zoom.
Let me say initially that I find it to be a very good lens, capable of rendering sharp, high-quality images, with beautiful color rendition and high contrast. But it has a few quicks, and it pays to understand how the lens works, in order to get the optimal image quality from it. The lens is optimized for shooting wide open, and has, for a zoom, really good image quality in the entire focal length range even at wide apertures. For me, I find it worth the money. Sharpness at the centre is on par with famous primes like the 50mm’s.
Let me also add that I cannot claim my findings to be universally applicable. However, I have tested two different samples of this lens that I believe are in good condition, one of them, which I own, both before and after a very thorough adjustment by Nikon (DFA A/S). I also think that my findings are consistent with the vast majority of images I have seen from this lens, and that the performance issues some are reporting to a large degree can be explained at least partially by the points below. In addition, my findings also appear pretty consistent with most comments given in Bjørn Rørslett’s review and that on Photozone, etc.
I have focused on testing generic and comparative patterns of behavior for this lens.
The most common complaint seems to be soft sides and corners at the 17-20mm range, even when stopped down. For some inexplainable reason, many of these reports are refuted by others that attribute the problems to user errors or use of AF outside its nominal application (even when AF clearly has not been used for the reported images).
I believe there can be, as usual, some “lemons” amongst the many samples out there. But I also think that part of the different reports are based on differences in usage pattern, slight sample variation within spec and different focusing techniques that all can contribute to a difference in image quality. Some lenses are rather forgiving, and will render almost everything sharply in focus by f/8 or f/11, no matter what focus point has been selected. This is not the case with the 17-55mm f/2.8!
All lens design is about finding a suitable compromise for the application. I don’t think the perfect lens exists yet. The price paid for the stellar wide open performance across the focal length range appears to be some quirks, described below.
If you don’t need the wide-open performance, get a lens with a different set of compromises. If you only shoot landscapes, you may have a slightly easier life with another, cheaper lens that is capable of rendering beautiful images by f/8-f/11, but maybe has a variable aperture or slightly inferior colour rendition or acutance.
Since example shots of this lens are often refuted by lack of technique, I have paid extra attention to all those details. The outdoor shots are done with the camera mounted on a tripod, using a remote release and mirror lock-up (MLU). To ensure that the plane of interest is parallel to the sensor, and that all buildings I use for the analyze are within the same flat plane, I have used Google Earth and GPS technology to ensure a proper positioning and alignment. For the indoor shots, I have used ISO standardized test methods and targets (slant edge).
I have not been relying on using AF; rather I have focus bracketed in two different ways: For the outdoor shots taken at the “infinity” setting, I have done a series of shots at varying apertures at each focus setting, and then refocused in small increments by manually adjusting the focus ring. For the slant edge test charts taken indoors (using a flash as illumination) I have moved the tripod in similar small increments, to vary the focus distance.
1) The lens is impressively sharp wide open at all focal lengths, which is what it is optimized for. It is excellent at the centre at f/4, and only marginally less sharp at f/2.8. Both lenses tested show the same kind of performance, with relatively minor differences. Before adjustment, my sample had a slight decentering, but after adjustement, it is now very even and is the sharpest by the two, albeit with a small margin.
2) There is quite some field curvature at the wide end, becoming practically negligible above 28-35mm.
In particular, this implies that A) if you focus wide open at a remote subject parallel to the image plane in such a way that the centre is as sharp as possible, the sides are not critically sharp. Alternatively, if you B) focus in a such a way that the sides are as sharp as possible, the centre will be focused “beyond infinity” and hence less sharp.
3) There is quite a lot of remaining (uncorrected or purposely designed) spherical aberrations that causes focus shift as you stop down, in particular at the wide end. This is somewhat visible at the centre, but much more pronounced at the sides and edges. I don’t know why this is so, but it may be related to the intended lens design and be part of why the lens is as sharp as it is wide open, especially at closer distances.
2) and 3) combined means this lens has a rather unorthodox behavior: If you focus at a distant subject with the centre sensor and compare the sides at f/2.8 and f/5.6 for distant objects, they get gradually worse until f/5.6, from which point they start to improve again. I don’t know of any other lens that does this.
Even for closer subjects, this appears to be true. The focus shift resulting from the spherical aberration causes the DOF zone to grow only on the side closer to yourself, rather than extending further away. In other words: It will not extend beyond the focus point until you get to f/8 or f/11, where resolution anyway starts to degrade due to diffraction. So you should never use hyperfocal techniques with this lens for optimal sharpness.
Rather it appears that the correct way of getting a sharp shot at infinity, both centre and at the sides, is to focus past the point of infinity, and then stop down. This will cause everything to become sharp. Using one of the outer sensors could be one way to get at least some of the way towards a better focus point, but it is not enough to get critial sharpness at f/8 at the edge.
Note that even if you focus beyond infinity and try to optimize the edges for sharpness, the impact stopped down on the center is not huge.
Details – Outdoors at “infinity”
For the outdoor shots, I have focus bracketed as described above, and shot a building complex that is parallel to the sensor.
The closest tested focus setting is usually labelled “A”, and the higher up in the alphabet, the closer you get to infinity, until reaching the end-stop. I tested this with two different lenses: Mine (labelled OWN) and one lens I currently have borrowed from Nikon/DFA (labelled DFA) for the purpose of doing comparative testing — my thanks to Nikon for assisting in this way to help figure out the root cause, and in general also for excellent service over the years and having spent the needed time re-adjusting my lens to its optimum performance.
The complete set of test images (278 raw images in total) can be made available upon request, but comparative 100% crops of 17mm, 24mm, 35mm and 55mm shots at f/2.8, f/4, f/5.6, f/8 and f/11 of the centre and a representative part of the right side are available below. These are all process in Capture NX, using low in-camera sharpening and a slight amount of USM to make it easier to see the differences by the unaided eye.
Comparing the shots from my own lens at 17mm at the centre shows that the optimal focus setting for f/2.8 is at “F” — which happens to be at the very end-stop of the focus ring. The lens simply cannot focus any further. Comparing the right side shows that the best focus is still at “F”, but it is also easily visible that the f/4 and f/5.6 shots are more blurred than the f/2.8 shot, and that you have to stop down to f/8-11 to recover the sharpness.
However, my lens clearly also shows its current problem, for which it will receive another adjustment soon: The back-flange distance at 17mm is not correctly adjusted, so the lens cannot focus past infinity, as it should be capable of doing. It also means the lens is far from being par-focal!
The same shots with the borrowed lens at the centre and side are worth an look also. At the centre, focus setting “G” is sharpest at f/2.8, but at f/5.6, it appears as if setting “H” is sharpest, and “I” is sharpest at f/8-11, although the differences are rather small. But the sides tell the story more clearly: Probably “I” is sharpest at f/2.8 (due to the field curvature), “J” or “K” are sharpest at f/4 and “L” is sharpest at f/5.6. “L” is by the way at the end stop for this particular lens. So the focus shift is very visible here. I have also, for your convenience, I am also attaching the key findings here — click to see it larger on top of this page:
Since my own lens exhibited the same behavior before adjustment, I am rather confident in saying that at least those tests I have done on multiple lenses show beyond reasonable doubt that this behavior is real. And if I compare the many samples found on the net, or those that people have kindly forwarded to me, shot under similar conditions, I would say that the simplest explanation that fits almost all of the results would be to assume that this behavior is inherent in the lens design, and not a result of poor QA or poor testing technique.
The shots for the same two lenses are also available at 24mm(OWN centre, right, DFA centre, right), 35mm (OWN centre, right, DFA centre, right) and 55mm (OWN centre, right, DFA centre, right). A study of this will show a slight amount of focus shift and field curvature more or less at all settings, but I’d claim it is negligible above 24mm.
Some people claim that they get better results by focusing at 55mm and then zooming out. One possible speculation is that their particular lenses are “almost parfocal” but will put the focus point a bit further out then the optimal setting at f/2.8 17mm.
Details – Indoors slant edge targets
First: My tests are not directly comparable to photozone tests, even if I’m using a somewhat similar methodology. The main reason is that they are not intended to be compatible, although I could possibly reproduce the same tests here. But the tests here are good for what they are: Comparative shots.
As described above, I have a series of shots (270 raw files) of a standard ISO test target, shot at varying apertures and varying distances. I then calculate the MTF 50% (uncorrected for any camera sharpening) from the Lightroom generated images at the centre and at the right side, and am then able to plot the MTF value versus apertures and distance.
The findings can be graphically depicted as follows:
The graph above now looks at my lens at 17mm. Inspecting the data behind the graph above, the point of best centre focus (highest MTF 50%) at f/2.8 is at distance “C+”, the yellow band (slightly further away than the “starting point”). At f/4, it is at red-bordeaux “A+” (which happens to be where the AF would put it most of the time), and at f/5.6 it is at the light-blue “C-” (slightly closer to the test target).
Here is then the same graph for the side:
This graph is rather telling.
For the sides, the best focus point at f/2.8 is apparently at the dark-blue “E+” (even further away than the best central focus at f/2.8(!)), and at f/4 it is at the light-blue “A-“. For f/5.6 at f/8, it is at the blueish-magenta “H-“, which is the closest tested point.
So it is very clear that the lens exhibits focus shift at the sides as you stopped down, in a rather pronounced way.
Furthermore, looking at the curves depicted here (and several more I also have made), it can be seen clearly that for objects farther away than the focus point, these will actually loose in sharpness/MTF 50% when you stop down. Only objects closer to the camera will increase in sharpness, and exhibit the normal “gets better as you stop down” syndrome.
Thus, a clear conclusion is that the focus shift is real and pronounced at the sides, causing a growing amount of field curvature as you stop down, even at close distances.
And just to rule out any errors, I have done the very same tests with a 50mm f/1.4 (and earlier also with 10 other lenses), and none of them exhibit this pattern. By the way: centre resolution of the 50mm f/1.4 at f/4 is about the same as for the 17-50mm f/2.8 at 50mm, but the side resolution is better.
Compared to the 17-35, the the 17-55 is sharper wide open, both at the centre and (a bit more pronounced) at the sides. One example can be seen here:
At 17mm the centre, the two are just about equal in performance stopped down. At the side, and without focusing past infinity, they are close in performance at f/4-f/5.6 for distant objects, with the 17-35mm having a slightly higher resolution and contrast the more you stop down, and beyond f/8 the 17-35mm is having the lead. However, at f/11, where diffraction kicks in, you have to look really close to find the difference.
Focusing past infinity brings them very very close, however, with a slight edge to the 17-35 for higher overall contrast. Note, however, that if you compare objects closer to yourself, the lead easily goes to the 17-55mm.
At 24mm: The lead still goes to the 17-55 wide open, and they have very similar performance stopped down, with the 17-35mm having slightly higher resolution at the sides, but not by much.
At 35mm: Clear advantage to the the 17-55mm at any f/stop, and it also exhibits higher contrast.
So the bottom line: Either lens will do the job. If you are only shooting f/8-f/11 at 17mm and care mostly about foliage at the horizon, pick the 17-35 (or maybe even better, the 12-24mm). If you care almost solely about wide open performance, pick the 17-55mm. Other than that, it is a wash.
Compared to the 18-70mm
At the wide end (17-20mm) and stopped down, the sides of the 18-70mm has slightly higher resolution unless unless you re-focus as described above. The centre is, however, clearly sharper for the 17-55mm. When focusing beyond infinity and stopping down, the 17-55mm has the clear lead. The 18-70mm’s I’ve tested all have a slight “smearing” of the picture, and appear with less “pop” than what the f/2.8 zooms give. One example is here (100% crop also):
At longer focal lengths, even though the 18-70mm is quite good, it cannot really keep up with the f/2.8 zooms, although the field curvature needs to be taken into account if you want the sides to be critically sharp.
Compared to 12-24mm
I’ve tested mine compared to the (Tokina) 12-24mm f/4. At 12-20mm, the 12-24mm is quite sharp stopped down a bit (and even wide open), loosing just a bit of resolution at f/4 wide open. For an overall landscape lens, it is hard to beat. It has a tad lower contrast and “pop” than the f/2.8 zooms, but is sharper stopped down and appear less smeared than the 18-70mm.
Compared to primes
Performance is overall somewhat similar to the 35mm f/2 or 50mm f/1.4 and f/1.8. Centre resolution is quite similar even at f/2.8, but the sides and edges on the 50mm f/1.4 is better at f/2.8 and above (the 50mm f/1.4 is reasonable at the centre at f/1.4, very good at the centre and good at the edges at f/2, and excellent centre and edges at f/2.8 and above, whereas the 50mm f/1.8 is not so strong at the edges and nor is the 35mm f/2).
Since writing the original article, my own 17-55mm (labelled “OWN” above) has had its back-flange distance adjusted and now properly focuses past infinity and thus behaves like the “DFA” sample above. It is still my “bread-and-butter” lens, but I regularly uses the “focus past infinity” technique for “infinity landscapes”.
Also, the newer 24-70mm f/2.8 lens has now been around for some time, and plenty of sample images are floating around on the net. Let me be clear on this: I do not have any direct experience with this lens. However, looking at other people’s evaluation of it, there may be some of evidence of it being designed with similar optical compromises. Thus, it does, to my interpretation of the results, appear to have a somewhat pronounced field curvature at the wide end, and I suspect it may also have some amount of residual spherical aberration, quite similar to the 17-55mm. If that is the case, I would not be surprised of Nikon sticking to a proven lens design methodology.
One write-up by Hendrik van der Veen shows some evidence of this, and also over at photozone.de, Klaus points at the same phenomena of field curvature. But please check the evidence for yourself before drawing any conclusions. Please note: It looks like this newer lens is generating similar kind of heated discussions as the 17-55mm did/does, and I have no experience with it. I’m just politely pointing out that it may have somewhat similar characteristics as the 17-55mm has. YMMV.
And surely both are wonderful lenses. At this point in time, it looks as you should get the 24-70mm if you have a FX body like the D3. But you have a DX body like the D300/D200/Dxx, you may be better off with the 17-55mm as a general purpose mid-range zoom (unless you mainly do portraiture or wide-angle landscapes). The image quality from both will be excellent.