...
Interesting premise that I am trying out now. I rebooted the machine again, lowered the o/c from +140/+350 to +125/+300 and restarted hsrminer. Reported speed is now around 1280; let's see if it holds.
Welp, that didn't work... Exactly 1 hour after restarting at a lower o/c the hashrate dropped to ~1150 (instead of ~1170).
I've just started poking around in event viewer but haven't found anything yet on driver crashes. Not sure where to look, actually, but I'll figure it out eventually... I'm sure.
EDIT - actually, backing off the o/c may, indeed, have worked. I decided to reboot the machine and try +125/+300 again and it has been running fairly steady at 1220 kH/s for 6 hours now. I never did find the gpu driver crash in the logs, but that isn't saying much.
Using nvidiainspector, I have a little script that runs in the background. It's got a 10min timer loop, and reapplies overclocks everytime it restarts. This way, if one card drops off for whatever reason, this both helps reviving it and makes sure overclocks are always current.
Happy to share it later tonight if you're interested.