2 things i suggest.
1. Make sure the TIM you are using is non-conductive. If you put any on the vrms,they can easily run over onto the pcb, i have had a card short because of this.
2. check the gpu die itself for cracks/breaks. Since AMD likes to just expose the dye, it can break easy as well.

If you sill have a dead card on you, i would just comb the pcb with a magnifying glass and see what is going on. There is most likely a common denominator between these problems that is being overlooked.

It is possible your mb is eating them. Have the cards all been in the same pcie slots?