looking at samples distance (sharp edge to edge the same between RigolDS4.png and RigolDS5.png) its not interpolation issue, its a fucked up averaging (if anything oversampled Peak detect) or sometype bit overflow or simply misplaced y value or whatever. i noticed this when downloading data to PC, i thought my PC app got bug, but it turned out from dso.
Well, the samples are spaced exactly as they are supposed to be at this sampling rate: 32 ns per sample:
(in this case a 10 MHz input signal was captured)
It's an aliasing issue all right, zoom out and you can clearly see it:
The same can be reproduced in Normal acquisition mode, it just needs the sampling rate to be even lower to manifest itself.
Here's a 10 MHz signal captured at 6.25 Msa/s in Normal mode:
I think Peak (or Average for that matter) acquisition just needs more samples per second to work properly compared to Normal. The issues related to the insufficient sampling rate simply manifest themselves earlier in acquisition modes other than Normal.
And it's clearly using linear interpolation, regardless of the acquisition mode, when acquisition is stopped and the captured buffer is zoomed in.