To clarify, the timestamp or t0 is actually taken from the system time when the Read is first called. The dt is calculated from the sample rate. There is more information on this in the KB below.
Why Is the Waveform Timing Information Returned by NI-DAQmx Incorrect?
<a href="http://" target="_blank" title="http://digital.ni.com/public.nsf/allkb/5D42CCB17A70A06686256DBA007C5EEA?OpenDocument">http://digital.ni.com/public.nsf/allkb/5D42CCB17A70A06686256DBA007C5EEA?OpenDocument</a>
There is currently no spec for the time difference between when the Read is called and when the first sample is actually taken. This will be difficult to spec because it would all depend on the enviroment, network traffic, distance from router, etc. When the task is started, it would send a message to the device telling it to initiate the acquisition and if the Read is called immediately after the Start Task is called the time difference should be minimal.
We do, however, know that the dt will be more precise because the time between when each sample is actually taken will be determined by the internal clock.
Sorry for not being able to provide a more exact answer, but it really depends on your setup and enviroment.
I was able to setup a quick test to give you a ballpark figure. To perform the test I set up a single point on-demand read. This means that exactly when the read is called it will send a message to the DAQ device, acquire one sample, and return the sample from the read VI. I used a Tick Count on either side of the read and took the difference to find the exact time it took to take one sample. This will give us a round-trip number, so really you could divide that by 2 to get an estimated time between when the task is started and the sample is actually taken.
With the device fairly close to the router and only one device connected it took around an average of .2 milliseconds round trip. However, increasing the distance from the router and adding more devices, meaning more network traffic, increased the time up to around 1 millisecond.
I have to reiterate one more time that this is no guarantee that you will see the same performance or numbers and it all depends on your setup and environment. Hopefully this will give you some sort of guesstimate on the approximate delay.