Comparing Measurement Computing USB 2416 vs NI9215

[LVNinja]

We're considering using cDaq for an application that  will read a voltage signal of approx 0.26 to 0.36 V. 

I did a test with a measurement computing USB 2416 -24 bits resolution and compared with NI 9214 4 16bit simultaneous AD.

For both converters I used the following parameters:
S/R - 500 S/s

Range -+10V (just because NI9215 only accepts that range so I used this in both for fear comparison)

Nbr of Samples - 2500 samples - 5 seconds of data. 

The results are shown in the screen shot bellow. 

I subtracted the mean from all the array and added all the "noise" together. For the sensor at 0.22V the sum was 

USB 2416 - 0.04

NI9215 - 13.673

The graphic shows the histogram distribution of the errors in percentage. 

I understand that the NI9215 may not the ideal for my application, my goal is to use a NI 9205 (32 Inputs 16 Bits with -+1 V) but my concern is about the noise level on this board. 

Can anyone from NI to give me some guidance if the NI9205 will have better performance ? Am I missing something ? 

I tried different sources, one precision power source at 0.7V and 5V signal from USB2416. at 5V they are equivalent but bellow 1V it looks very noisy. 

I tried to borrow NI cards for test but it's not available. 

I appreciate all inputs/comments.