So what's the culprit here? Looks like you have limited power supply, if reducing the voltage works then it's the power, I'm glad you are able to figure things out, can you share the exact watt that your GPUs are pulling? Thanks
Memory: 2100@875mV
Core: 1050@875mV
31.2mh and 97w
I have 6 identical gpu, three of them turn unstable after 3 hours with that config, so I have reduce them its memory frecuency to 2050 reducing its performance to 30.7mh and 94w.