I did this test a few years ago and thought I run it again. Mostly to see if the new drivers for my Tascam us1641 are really as good as I was experiencing. I was right. In my test prior to the new drivers the Tascam was off time. Now it is bang on, but,, only in ASIO mode. This is why I'm posting this thread here. I think everyone needs to do this test. Especially if you've been advised to use WDM mode.  Don't unless you test!   
Not all interfaces will pass the test under WDM mode. Very few probably. 
 
And it is very clear why MME mode is to be avoided for anything but playback. 
 
 I ran the test using 3 different computers, My main DAW ( W10) , Laptop (W7) and my old DAW (W7) .  All are 64 bit windows up to date.  
Main DAW and Laptop running Sonar Plat latest build ( my signature) Old DAW Sonar studio x3e 
 
It was interesting that my Scarlett 6i6 will not run on any computer in WDM or MME mode. I can change to these modes and run the wave profiler, but the minute focus is on the project the Audio Engine stops.. and at any setting or buffers. The Tascam must have WDM drivers or the Scarletts drivers are not stable in this mode.
 
The test results were exactly the same on all 3 computers 
ASIO mode bang on ( even if you zoom way in.. and WDM mode a little late,, MME mode WAY late. 
 

 
 
FYI The loopback test is easy.
First set up you interface with a short cable running from an output to an input. 
I unplug my monitors from 1/2 to avoid mishaps and used output #1 into input #1.  
You can use headphones if you want to listen,, but your eyes are more important. Turn them way down. 
Turn the input gain off for now. 
 
Open a blank Project.
insert a MIDI track and draw a few snare or kick notes. Only need a few.
Then insert an Audio track for each interface or mode you want to test. 
Insert any drum synth and freeze it. This creates an audio track.
Notice it should line up with the midi track.
Assign the synth audio output to MASTER or which ever output your going to send from. 
For the audio tracks other than the synth set input to match your patching. 
Then set output to "none" to avoid feedback loop. do not engage INPUT ECHO. ( double insurance) 
 
Now pick an audio track, name it and set it to REC. 
Hit play and adjust your level at the interface. Try for a good almost over the top level.
Now record the track.  
Repeat for as many modes and interfaces as desired.  
 
I tested using different buffer settings and this had absolutely no bearing on the results. So just use what ever you normally use. I used 44.1/24 .  
 
 

I did the same test a while back, 6i6 and asio, and saw same results!!!

Here's a few screen shots from when I did the test a few years ago testing a few different interfaces I have kicking around.
This is a Behringer UCS 200 , I came with mixer. They have a few versons of this and possibly the cheapest interface on the market.. But no ASIO drivers,, it uses what it calls USB audio codex and Sonar only works in MME mode.I use it for live music playback without issue,, it's dead simple plug and play any OS. Always works... But you can see why it's a bad Idea for trying to overdub. 

 
 
This is this cool old PCI card my son scored from a Radio Station rig. The company still supports it even though it's from I believe 2002.  And even though I mislabelled it,, the second track is WDM mode and it surprisingly is bang on. 
 

 
I also tested my old Creative Labs Audigy II card. Interesting that it was early in WDM mode,, it was close to right on in ASIO. I don't have the screen shot stored, but it drifted over time in both modes. 
 

 

cool stuff Johnny. really interesting tests. going to give this a try myself.
thanks for sharing.

It's more for people with iffy interfaces than those of us who purchased something known to have good drivers. 
But I was pleased to see that Tascam did improve my drivers for the us1641,, to bad it was after the fact and I'd already bought the Scarlett. But it's good to have both. 

I own a Scarlett as well (The 18I8)
some really cool tests that I enjoyed reading. And good info for some who have older AI units.

Maybe I am reading the screenshots wrong, but it looks like you are "recording" the looped back audio earlier in the timeline than the original in some tests. If so, then the problem must be some sort of delay compensation that is overcompensating. A too large buffer or processing delay would result in the looped back signal being recorded after the original. If the compensation is consistent, you might be able to correct it by setting the delay manually.

Good timing on this. I recently realized that I have to nudge my takes forward using 1 medium and 1 small nudge. Total headache. If I run this test I assume that I use the results to apply a manual compensation?

slartabartfast
Maybe I am reading the screenshots wrong, but it looks like you are "recording" the looped back audio earlier in the timeline than the original in some tests. If so, then the problem must be some sort of delay compensation that is overcompensating. A too large buffer or processing delay would result in the looped back signal being recorded after the original. If the compensation is consistent, you might be able to correct it by setting the delay manually.

 
This is why it's important to perform this test. Some audio drivers "lie" to Sonar and the offset is calculated incorrectly. The loop back test will show you if your driver is performing correctly, Sonar can only calculate based on what the driver reports. 
 
And as I said buffers have no bearing on the results. You can use your lowest and highest buffer and the results will not change other that run the risk of an audio engine dropout at the lower settings. 
 
Of course for the test we use a brand new project with absolutely no effects. Your welcome to try the test and see what happens when you add a "look ahead" effect to your master bin. 

Thank you, Johnny, for the step-by-step. I've wondered about this procedure literally for years, and haven't gotten around to figuring out how to do it. I have a first gen 6i6 and I do a lot of guitar playing through amp sims, monitoring back through the SONAR input. Latency is acceptable but noticeable. Maybe now I can find out what it really is. (Latency is supposed to be better with 2nd gen Scarletts. I wonder...)