BobF
Is it just me?  I don't see 17ms RTL with 128 sample buffer to be that bad.  I'm happy as can be with 8ms RTL @ 64 samples.
 

 
Additional round-trip latency on top of 2x the buffer is typically just a millisecond for A/D/A conversion (20-30 samples each way) plus whatever the bus adds - maybe 1-2 milliseconds for PCI(e), and anywhere from 3-6 for USB. So I would say your 8ms at 64 samples is on the high side of normal for a USB interface. 17ms at 128 samples is pretty far out of the normal range.
 

To add clarity. The CEntrance measurement provided above was performed with Sonar closed.  What I was trying indicate in the post above as that I also see a reported round-trip latency of 16.19 ms inside Sonar if i don't have a project open.  If I open a project I see a reported round-trip latency of 17.7 ms. So it appears that just opening a project increases the round-trip by 1.51 ms (which I guess could be a result of of some of the plug-ins in that particular project but I am not sure). 

What happens if you uncheck 64 bit precision engine?
Also your timing is octacapture 1-2 which is correct but why is the record latency adjust set to octacapture 5-6? Try making that octacapture 1-2

Just an idea

brundlefly

lfm
Just make a analog loopback on interface - and record metronome on that track.
If they line up reported is correct.

CEntrance does the measurement for you, and is sample-accurate without exception in my experience.

That is good to know.
But also good to know what Sonar get as reported and where it places recorded clip - there was a deviation in reported samples.
Not end of story - but just a clue how to proceed.