The LTZ1000 circuit is about 5 times less noisy, and about 20 times more stable (1 year, 45°C oven) as the LM399.
Hello,
from my experience the long term stability of LM399 is also in the ballpark of 2 ppm/year after 5-9 kHrs run in time.
So not a factor 20 against LTZ1000.
In the picture: ADC13 (around -1ppm/year) measuring different devices.
So "real ageing" has to be corrected by -1 ppm/year.
LM399 CH6 + CH7 are from "day zero" on.
Other sources "well aged".
LTZ#1 had a short on day ~860 (-1.9 ppm jump + increased ageing of green curve).
You never should operate a LTZ1000 without buffering. The LM399 is insensitive to shorts against ground.
LM399#2 had been accidently operated several days upside down "pins up" on day ~920.
This created a increased ageing. (red curve).
So keep the LM399 always in upright position.
All in all the short term noise of a LM399 is much larger.
But long term stability is usually (except of some "stinkers") not much more than a LTZ1000.
If memory serves noise is typically not presented as voltage or current but rather as noise power, how does summation of similar noise power voltages eliminate noise?
The only way is averaging several sources.
Doubling the number of sources reduces (in theory) the noise by -3dB.
Practically the noise of a LMx99 has a large stray (from device to device) compared to a LTZ1000.
So I would recommend to select your LMx99 by noise.
There are several measurement amplifiers to do this discussed here in the forum.
With best regards
Andreas