Having a few moments of Zen with three generations of gear. The 34401A and 735A have been locked together and on for over a month and I haven't used the 34461A for over a week (yes I am a bad owner).
This is why you don't just flick on a device and expect it to read properly. It took nearly 30 minutes before the upward drift on the 34461A stopped and got somewhere near steady, stable and accurate. Sitting about 5PPM low against the 34401A on a very mild 15C day if it warms up toward 20C that gap will close towards 1-2 PPM as the 34461A isn't as stable.
My 34661A takes about 15 minutes to fully warm up for all but the most critical measurements - that equates to a 11.2 uV upward drift of the measured zero of a short over 15 minutes on the 100 mV range. It takes another 15 minutes for the final ~1 uV drift up to a settled true zero. So mine's about -120 ppm off at turn on, hits ~ -1 ppm warm-up error in 15 minutes, and finally reaches 'spec' at 30 minutes.
It's all a question of what kind of measurements you're firing it up for. Need ~ 1mV accuracy on the 10V range (100 ppm/0.01%)? No significant warm-up needed. Want to squeeze out every last PPM like you were, yeah you'll be waiting 30 minutes. If I want to make critical measurements I power it up with a short on the input, turn the trend chart on, go and make a coffee and don't touch the meter until all the warm-up drift has fully settled out.
Because of the sheer measurement power of something like the 34661A it's easy to get spooked (in everyday non-critical measurements) by what are really quite small changes. Case in point of me getting nearly caught out:- I was measuring the V
f of a 1N4002 on mine the other day. I already had the meter on displaying trend with autoscale, intending to check a few diodes in quick succession just to get a quick visual indication of the diode to diode V
f - mere curiosity about the batch consistency of a cheap commodity component.
Measured the first diode and grabbed the diode to change it for the next. Wham! Trace starts shooting down the screen. It took me a half minute to register that the downward slope was the -2mV/C tempco of the diode and what looked like a large change was down to the fact that the range initially displayed on the trend chart was only a few hundreds of uV. Final result when I'd corrected my careless handling and allowed some temperature settling time on each diode - less than 1 mV diode to diode variation for the first few diodes on the same bandolier.