Agree, as the nonce is like randomly located between 0 to unit32_max.
when miner divide the work to each gpu,
does it divide in a way that they won't overlap?
Most likely. What the point of computing twice the same thing by the way?
I have a gpu miner with mixed gpu. 1080, 1060 6GB and 1060 3GB.
if miner divide the nonce evenly to each gpu, the chance of finding the nonce will be also the same for each gpu.
however, i can see 1080 finds more share than 1060. as 1080 is much faster.
I suspect each thread are trying the nonce randomly.