Comment Re:Um... (Score 5, Interesting) 187
Given that the particular hardware setup is detailed here (a GTX 480 achieves the 1 billion keys/sec figure), and the algorithm used (radix sort) has known asymptotic behavior (O(nk) for n keys of length k), 10^9 keys/sec is quite meaningful, particularly since it's a significant implementation challenge (possibly even an algorithmic challenge) to port this algorithm to a GPU.
Furthermore, I think sorting speed is appropriately measured in keys/sec. Big-O does not in fact describe the speed, but rather the upper bound of the growth of an algorithm's asymptotic running time, which needs to be paired with the implementation, architecture, and data set to determine a speed. It turns out the constant factors can actually be quite important in practice.