https://i.imgur.com/lj274W7l.jpg
It is very well possible these were 'NOPOPs',
parts that were not ment to be populated
on the board.
Sometimes after testing a design, it turns
out some parts can be omitted without
decreasing the performance of the circuit.
Parts are then not placed during production
in order to save a few bucks.
When the solder on the pads looks nice and
regular, there were never caps on that location.
I don't think they should be DNPs, as they're the output caps for the buck regulator (the same part (TPS5335) you used on the Bitfury 25GH boards).
If anyone's interested here's the relevant schematic section.
Edgar: from what I've seen the heatsink compound used is of passable quality - the actual application during assembly varied wildly (too much, too little). The HS screws were also installed erratically, resulting in voids under the heatsink where no or little contact was made. I also had several where cutting/tapping fluid (which hadn't been cleaned out properly) oozed out from the heatsink screw holes and mixed with the compound, resulting in a runny mess.
Don't get me started on the HS design itself..
comparing the designs, the Bitfury design only has 500uF in bulk caps(ceramics), while bitmain rev.1 has 3000uF in bulk caps(electrolytic) and Rev.2 has 1500uF
the purpose of the caps at the output is to smooth the voltage to as much of a flat line as possible by providing a resevoir to fill the valleys in the mVp-p