Hi Bart, apparently the 5Ds raw is also available. This is the link to the 5DsR raw just in case.
Hi Samuel,
As you've noticed it's a set that was shot with a different lens, not a comparison between the two bodies with the same OTUS lens. But the differences at f/5.6 are not that large.
I expect that the 5dS/5DS R with a good lens will peak at f/4.5 or f/5.0, because of diffraction, the OTUS may score best at a tad wider but not much, if any. No matter how good a lens is, there will be residual aberrations which get more destructive as more of the lens glass edge rays are involved in image formation. It's just that difficult to achieve perfection. Also, while lens resolution can be higher at even wider apertures due to lower diffraction, the sensel apertures on the sensor chip itself will act as a physical averaging filter. Smaller lens projected detail will not be imaged as higher resolution, just generate more aliasing artifacts.
So at that f/4.5 aperture it would possibly show the most resolution difference between the bodies if there is any. The diffraction pattern diameter at f/4.5 is 6.25571 micron in diameter for 555nm wavelengths when an 85mm lens is focused at 3 metres, or 1.51x the sensel pitch of the 5DS/5DS R. In my experience with other cameras, with the diffraction pattern diameter at 1.5x the sensel pitch, I start seeing some deterioration in the micro-contrast, nothing alarming, but the decline will start to increase with narrower apertures beyond that. Also my actual lens tests (although with different lenses and bodies) show that in practice the better lenses do peak near that 1.5x sensel pitch diffraction diameter pattern turning point. Lower quality lenses require a slightly narrower aperture to eliminate a bit more residual lens aberration before diffraction becomes a visibly dominating factor.
The resolution test files at f/5.6 will therefore already, in my estimation, have a bit of a leveling effect between the two bodies from diffraction. The difference in aliasing effects will thus be slightly less pronounced at f/5.6 compared to f/4.5. But this would also require extremely difficult to achieve perfect focus to become visible with some level of accuracy. At f/5.6 the DOF zone at e.g. 3 metres is 56.7 mm (slightly more than 2.2 inches), but within that zone there is only one very narrow plane with really perfect focus. To hit that perfect focus plane is hard and time consuming, it would require shooting a large number of different shots with very slightly different focus distances of the camera along that DOF zone range, e.g. with a Stack-shot rail or similar. So there may be small sub-perfect focus settings in the test shots from DPreview that also add some deterioration for one or the other body.
Do you think in this particular case, 5Ds vs 5DsR, that OLPFs are still relevant? My personal preference after attempting to process both raw files in Raw Therapee with AMaZE demosaicing is that yes, it is. But it would be great to learn from your own vast knowledge and experience.
In general, and in particular with Bayer CFA limitations that will create false color artifacts,
proper low-pass filtration is to be preferred in any Digital Signal Processing (DSP) scenario. The lenses will resolve more detail than the sensor can, and the Bayer CFA will therefore by definition generate false color artifacts, unless the Raw converter is very clever in hiding them (but hiding them is necessary). I've understood from comments by Emil Martinec, that the final steps of the AMAZE demosaicing algorithm, the 'ZE' part of the acronym, finishes the conversion with
Zipper artifact
Elimination, which actively
adds a bit of blur, which e.g. the IGV algorithm doesn't. That's why that IGV method may be locally sharper, but with a higher risk of creating zipper artifacts on edge detail. In my experience AMAZE strikes a very good balance between resolution and yet low risk of generating artifacts.
However, although (optical) low-pass filtration before sampling is preferred, it does need to be weighted against the quality of the optical signal. So if one
never shoots at the optimal aperture, then the diffraction (or residual lens aberrations) may have enough of an effect to equalize the differences between the bodies. We are getting smaller differences between OLP filtered versus non-OLPF cameras with shrinking sampling pitches, but at 4.14 micron there is still a difference when optimal conditions (lens correction, focus, and aperture, filterstack, microlenses) are present.
The question then becomes
how much of a difference are we able to see (in optimal and in adverse conditions), and is it big enough to be concerned about (and pay extra for).
I will have to do the conversions of the f/1.8 lens Raws at f/5.6 myself, before I can judge the significance of the differences in (presumed) perfect focus circumstances. I may have a bit of time tomorrow to do a conversion of the Raws and do some measurements.
Cheers,
Bart