The last thing I want to show is that there is another type of artifacting that can occur in situations where you push the secondary ISO very high, for this example 3200, and you have thin lines (i.e. small branches or wires) running almost parallel to the sensor rows.
In theory, this can happen at any iso combination. The way the cr2hdr postprocessing utility works, it tries to interpolate the scanlines and reconstruct missing data in the the shadows and highlights that is only available in one half-picture. Well, if detail happens to be just in one scanline and is clipped, that's that. But Alex (the lead ML programmer) might refine cr2hdr even more to minimize the inherent problems.
But compared to the difference between a vanilla 5D3 and the A7 and this becomes a pretty amazing result.
Thanks for the comparison! Fyi, the 6d works even better with dual iso since the dynamic range is a bit higher and it has less banding noise when pulling the shadows. I use dual_iso more and more at iso 100/800 these days, gaining +2.5ev dynamic range with nearly no iq drawback on the 6d.
Unfortunately, in real life you cannot reduce the comparison to tech aspects. The workflow problems of dual_iso vs. a native high dr sensor are in my experience:
- it's harder to check for focus in camera because the image is interlaced
- it's impossible to check for colors because they are screwed before processing with cr2hdr
- postprocessing hassle, esp. time required for cr2hdr processing
- dual file storage because you want to keep around the original cr2 in case cr2hdr receives further improvements
- results usually needs manual wb (esp. tint) setting even though cr2hdr tries to autodetect it