The issue with the constant reboots of systems with a lot of GPUs seems to be related to single cards. As overclock apply to whole machine you could have a single card/cards that can't handle the clock and hung means reboot of machine. Here is a screenshot of two 1:1 machines with same HW, 1st one reboot every 1hr, 2nd one is not. So my suggestion to keep the logs files "current minus 1" will help to see which card to underclock. Can we put clock values with commas to separate for different GPU: like 100,150,80,200,100.... ?

https://preview.ibb.co/c7gYyw/11.png