DC signal SNR on Analog In depend on collection time not on no. of samples averaged?

[Muni]

I am using M series PCI- 6221 board together with BNC-2110 box and LabView 7.0 sw. I have a DC signal
from a sensor connected to an Analog Input. I am calculating SNR of the signal using a series of data
collected by my LabView application. If 6221 is doing just sampling the SNR should depend
only on the number of samples I average. But instead it seems the SNR depend on the sample rate.
Example:
If my sampling rate is 100kHz and average 100 samples (i.e. over 1 msec)and save 100 of those averages to calculate
SNR I get 1000:1. If my sampling rate is 50kHz and still average 100 samples (i.e. over 2 msec)and save 100 of those averages I get a SNR of 1400:1. That is an improvement of sqrt(2).
On the other hand at 50kHz if I average only 50 samples (i.e. 1 msec)SNR is 1000:1, same as 100kHz with 100 sample
average.
This suggest that 6221 is doing its own time averaging in between sampling. Could you please explain, is there
anything wrong with my logic or am I doing something wrong?

[rpursley8]

This is the way it works. There is noise on every point you sample no matter what your sampling rate it. SNR drops according to how much 'time' you sample, not how many points. 10,000 points in 100 ms should give you close to the same SNR as 5,000 in 100 ms but 10,000 points in 200 ms will improve SNR by sqrt(2).

[Muni]

Thanks Randall, but what I am observing is that reading 5000 samples in 2 ms give
me sqrt(2) improvement compared to 5000 samples in 1 ms. How come just waiting longer
without doing anything extra, improve SNR?

[rpursley8]

Can you explain how you are doing your acquisitions and/or post your code? Are you getting a noise only signal and then sampling your sensor signal and determining the SNR from that or are you sampling a waveform and averaging the waveform several times before performing an FFT and then measuring the SNR?