Concerning intensity: There is no such thing like intensity which I'm setting to make the GPUs run faster, if this is what you're thinking about. The performance depends mainly on how well you've designed your algorithm and how well the program that implements it is able to utilize the GPU.
If I understood you right, zm fails on cudaMemcpy on some of your rigs. Does a particular GPU fail or do they fail all at once?
Most of the time it's GPU 0,1,2 that fail (probably because the load is distributed sequentially across all the GPU's, so the miner crash even before loading up all the gpu's), but it's across a whole 10 rigs of 6x1070, also to note they work perfectly on EWBF, hence why i was thinking ZM may push them harder than they can handle, but if nothing can be done "intensity" wise i guess i have to abandon the idea to use your miner on those specific rigs, and only keep the EVGA's one using it... It's also weird that even with no overclock the miner crash as fast than with OC.