For 32bit ints that in the order of 3.6e88 TBs.... To put that into perspective the visible universe contains an estimated of 10^82 atoms at most, so if a singl

For 32bit ints that in the order of 3.6e88 TBs.... To put that into perspective the visible universe contains an estimated of 10^82 atoms at most, so if a single atom contains 1 petabyte of data, and you want to sort all atoms and all or their data - this ML approach would be faster. I wish they had that discussion in the paper.