In the not too distant past I did an ISO-exposure test on my 40D, the full details of which can be found
Here.
My reasoning was that as I take several hours of lights, then darks, other object lights & more darks, the best I could hope for was some before (warmer) & some after (cooler) darks that didn't exactly match my lights.
My thoughts were to minimise the noise in the first place so any difference was also minimised, but I digress...
These values are "rough & ready" so should be taken with a grain of salt, especially in the low ISO region.
Below is the mosaic of dark frames & the table of average values, this seems to agree that for a given signal/exposure SNR will increase with increasing ISO, at least for thermal noise.
Attachment 73517 mosaic.
Attachment 73519 average values.
If we take for example 1min exp @ISO800 given a signal value of 16 & noise at 16 this would give SNR of 1:1.
At ISO1600 signal is amplified to 32 with noise at 22, or SNR of 1.45:1 & finally for ISO3200 SNR increases to 2.2:1.
That's all well & good but in my meager experience 5 min subs at ISO1600 or 3200 have horrendous noise even after dark subtraction (uncooled, tropical temps remember).
So looking at stacking an equivalent, same exposure or true signal (No of photons)...
2x 1min, ISO1600 - SNR 32/22 or 1.45:1 each frame.
1x 2min,ISO800 - SNR 32/18 or 1.78:1.
Ignoring any dark subtraction & I know this is a very simple case of only two frames but an "average" stack would not change the SNR at all in the ISO1600 case.
Even further if the signal was faint (as many astro sources are) & only supplied one photon per 2 min interval then the SNR of the average stacked 1600's would be halved as only one would contain signal.
This is an over simplified & extreme example but you can see where I'm going with it...
I'll agree that for a single, given length exposure that high ISO has greater SNR & that using as high as you can for a given exposure is good but for faint, stacked astro work...lower & longer still gets my vote.
Feel free to set me straight on this, always willing to learn something new in this area.
Justin.