Nikon Focus Consistency Wars: DSLR versus Mirrorless

I have known since the first day I got my Nikon Z9 that it focused better than any Nikon DSLR I have ever owned. When I got the Nikon Z8, I noted that it seems to focus exactly like the Z9.

I was a long-time holdout against mirrorless cameras, both because of their short battery life and their sluggish viewfinders. I didn’t see any particular advantage over DSLRs to persuade me to stop using my D850 or D500. Nikon eventually solved all of the mirrorless shortcomings that I knew about, and then some. I didn’t know just how much they had improved the focus system compared to DSLRs.

Instead of just ‘feeling’ like they focus more consistently, I did some tests to get some actual numbers comparing one of my mirrorless Nikons (the Z8) against my best DSLR: the Nikon D850. The camera sensors have the same resolution, so any measurements can be compared directly.

To eliminate any doubts about the lens being a limiting factor, I decided to use my 500mm f/5.6 PF Nikkor, which I consider my best lens, for the tests. Both cameras used the same ISO and exposure. I used a wired shutter release, and the D850 shots were done in “mirror-up” mode.  I manually de-focused the 500mm lens between each test shot, to force the cameras to do a full re-focus using AF-C.

My benchmark to evaluate focus accuracy is the lens resolution measurement, since higher resolution measurements correlate perfectly with accurate focus. I used a precision razor blade edge as my focus target, since it always gives me either equivalent or superior resolution measurements compared to my printed targets. I use the MTFMapper program to evaluate my focus target photos.

I tested both cameras using phase-detect 3D focus. I’m not interested in comparing slow contrast-detect focus, even if it might be more accurate. My D850 has been autofocus fine-tuned with the 500mm lens; my Z cameras don’t seem to need this crutch, although they include that option in their menus.

I took many raw-format photos of a precision razor blade that is in silhouette. I don’t want to use any automatic sharpening (such as when shooting jpeg) to affect the resolution measurements. The camera/lens is on a heavy support, so that each shot has the blade silhouette in exactly the same location. Make sure that the razor blade is perfectly parallel to the camera sensor.

Mirrorless Z8 versus DSLR D850

The focus subject is a slanted high-precision edge

The shot above shows the MTFMapper dialog used to perform “manual edge selection”. I have a razor blade with strong back-lighting as a target, and I put the camera focus point over the center of the razor’s sharpened edge. After using the computer mouse to select the portion of the blade image to use, I let the MTFMapper evaluate this edge to determine the lens resolution at this position and orientation.

The MTFMapper program lets me select all of the test photos at once; it assumes the target in each shot is at exactly the same position. I make sure that “reuse ROIs” is selected, and then click the “Accept Queued” button to have the program evaluate all of the photos without further intervention by me.

A typical shot after MTFMapper has finished evaluation

You can see a little cyan-colored MTF50 resolution measurement superimposed over the location on the blade where I requested the program to evaluate the resolution. In this shot, the MTF50 resolution measurement is 84.8 lp/mm.

I put all of the measurement results from MTFMapper into an Excel spreadsheet. I then plotted the resolution measurements against the camera filename (frame number).

D850 resolution variation with 500mm f/5.6 PF, AF_tune -4

The plot above shows the variation in resolution measurements for the Nikon D850. Again, I de-focused the 500mm lens and then refocused between each shot, to force the camera to perform a new phase-detect, AF-C focus in 3D-tracking mode.

The D850 results got an average MTF50 resolution of 63.2 lp/mm, with a peak reading of 81 lp/mm and the standard deviation (spread of measurements) was 8.12.

Since there was a single measurement higher than the others, the autofocus fine-tune setting could possibly be re-adjusted to get overall better focus, although the photo EXIF data in every frame indicates the same 4.47m focus distance. The spread of the measurements isn’t affected by the autofocus fine-tune setting.

Bear in mind that resolutions above 40 lp/mm will all appear pretty sharp, so these results are still good. Nonetheless, I decided to change the AF-tune and re-try the shots.

D850 resolution variation with 500mm f/5.6 PF, AF_tune -2

The D850 with the new AF-tune -2 results got an average MTF50 resolution of 71.7 lp/mm, with a peak reading of 84.8 lp/mm and the standard deviation (spread of measurements) was 10.3.

The slight AF-tune adjustment change definitely helped! Notice that there is sort of a split in the data. When I was shooting the target, I would de-focus the lens by alternating too near and then too far shot-to-shot. The camera seems to have a bias in focusing based upon the direction to achieve focus.

I don’t know if something might have shifted or drifted over time with the focus calibration. It was definitely worth re-checking this, since it resulted in a typical gain of about 13 percent resolution.

Z8 resolution variation using 500mm f/5.6 PF

The plot above shows the variation in resolution measurements for the Nikon Z8. Again, I de-focused the 500mm lens and then refocused between each shot, to force the camera to perform a new phase-detect, AF-C focus in 3D mode.

The Z8 results got an average MTF50 resolution of 74.0 lp/mm, with a peak reading of 98.2 lp/mm and the standard deviation (spread of measurements) was 6.7.  With the exception of the single reading of 98.2, the data was much more tightly grouped than the D850 results. I think that the 98.2 reading is some kind of a fluke which should probably be ignored.

If you’re interested, the average resolution of 74 lp/mm is equivalent to 3537 lines per picture height. I prefer using ‘lp/mm’ units, since it applies to any size sensor.

I didn’t notice any bias in resolution related to focusing from near-to-far versus far-to-near, either. The Z8 (and Z9) are just smarter about focus.

Summary

The Nikon Z8 (along with its big brother the Z9) has noticeably better focus consistency than the D850. After seeing how the D850 results dramatically improved with such a small AF-tune adjustment, I should probably explore using that option on my Z cameras, too.

Before testing with my Z8/Z9 cameras, I would have said that my Nikkor 500mm f/5.6 PF lens had a typical center resolution of about 63 lp/mm when using 3D continuous autofocus on my D850. After studying those results and modifying the autofocus fine-tune, I changed my mind to conclude that the resolution was closer to 71 lp/mm. Now, I can say that it has a typical center resolution of roughly 74 lp/mm when using 3D continuous autofocus on the Z8/Z9 cameras.

If you only shoot using slow contrast-detect focus (via Live View on the D850) then these focus (resolution) differences will largely disappear. If your lens has focus-shift issues when changing the aperture, then the focus variation differences between DSLRs and mirrorless cameras will increase. The Nikon mirrorless cameras focus at the shooting aperture when using phase-detect (through f/5.6), but the DSLRs don’t.

You need to shoot many tests and study those results to guide yourself toward optimal calibration, in the case of DSLRs. Statistics exist for a reason; you can’t really know how your gear performs without doing lots of testing. It’s possible that your mirrorless Nikon could squeeze a bit more sharpness out of a lens with autofocus fine-tuning, as well.

Mirrorless definitely wins the war of consistency over DSLRs when using phase-detect focus.