I've read the manual. But I can't find any information which would confirm my measurements. Quite the contrary: Page 74:
"But when performing a digital loopback test no latency/offset can be detected."

Could you or somebody else please explain?

Thank you

There's also the 32-sample "Safety Buffer" that came about in driver version 3, but this still doesn't account for the extra 33 sample discrepancy in your measured 199 samples (64+64+32+6 = 166 Samples).

Could RTL be mis-reporting somehow?  What does your DAW show in this case (record the loopback to another channel and see how it lines up to the original)?  The DAW is probably close to sample accurate - I'm not sure if RME's driver also takes into account the ADAT Tx/Rx buffer, but I assume it must.

Thanks Randyman & Matthias. Now it's clear - so now there is only 1 unidentified sample left : )

Like I mentioned before I would like to use RayDat with the UCX. In my case the calculation would be:

199 samples
+
14 samples for AD and 7 samples for DA (on all sample rates)
=
220 samples

Am I right ? Does RayDat and UCX in combination have the lowest latency of the RME products ? (at 96 kHz)

How about running the FF 800 as stand-alone ADAT converter with the RayDat ?
Does it add some internal latency too (besides the higher latency of the older AD/DA converters compared to ADI-8 DS MKIII) ?

Thanks

The 800 does not add latency in stand-alone mode, but due to the older converters the advantage is near zero compared to the UCX. Not sure what all this is good for...