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?