On the GPU, from that array of 48 integers I want to select every possible combination of 5 unique numbers. Each thread on the GPU should map to one specific combination. That is 1,712,304 combinations (or 1,712,304 threads). The order of the numbers is irrelevant, so I do not want duplicates. In other words, if I use the combination 0 1 2 3 4 5, then 0 1 2 3 5 4 should not be used because we already hit that case (see