Update:
I did get a timeout on the AMD GPU about 5 1/2 hours after a reboot. Cards crashed (still showed as hashing for about 4 minutes, but the calculated Hashrate/W jumped up):
Try to run all the AMD cards in once instance, and the nvidia cards in one instance of the miner.
------------
TBM 100% LHR unlock is coming soon.. My testcard's core clock is dropping to 900 when I boost the memory clock, and this is hurting the performance.
I read that others have the same problem, but perhaps I can tweak the kernel to use lower memory temps.
edit:
Stable test. Will let it run over night.

Shitty clocks on my 3080ti card because it trottle, but managed to tune it abit better. Kernel 0 with 9991 intensity.
If core drops when you raise the mem clock - you run into the powerlimit - increase it.
I have seen similiar things on 1 of my 2060S, all of this cards run at 45mh, on 1 card, if i raise the mem - the coreclocks drops to 750. For me it makes no sense because the card only use around 120w, but the powerlimit in the bios is a little bit lower than all other cards and i can´t increase the pl over 100% it is kept there.