We need to shorten the time between when a card 'times out' and when the miner restarts. I've never seen a card recover from a timeout and often times 3-4 will all timeout at the same time which drops my average hashrate significantly, which also affects my poolside share of a given round.
I mean ideally the AMD cards wouldn't just randomly time out, but if they do, I'd rather just have the process restart than wait for the inevitable crash while hashing at a low rate.
After a lot of head-scratching and tuning with many crashes and poor hash rates, I finally have found a combination of clocks and intensity that allows the 3090Ti to perform well.
WhiteFang is the 3090Ti and it has been running for about 10 hours now without any crashes.