I have the same problem but my rig consists of 8x 580 Nitro+.
I have a constant core clock on all of them (1200), set in Claymore config.txt as well as cvddc and mvddc of 857mV. But I have varied memory clocks starting at 2025 and goes up to 2100.
The main problem is that only one card crashes. Good thing the -minspeed argument is working and it restarts my miner should one card crash (together with -wd 0) and everything goes back to hashing a total of 244 Mh/s again. This is dual-mining by the way.
I wanted this issue to be resolved although judging by other's experiences, I'm on the lucky side as my rig still mines and restarts on it's own.
P.S. I used to have an 8x Gaming X 580 rig and it doesn't have this issue. Same board and components.
P.P.S It is not a specific card that is crashing. 4 out of 8 cards takes their turn crashing, one at a time.
Thoughts?
What is powersupply rating? Dual mining is power hungry, ensure that your psu can handle and supply power for all gpus, would suggest psu to have at least 10-15% allotted extra power from the wall.
And i've fixed the issue yesterday by updating to win10 fcu and 17.10.3 driver.