LXM: Better Splittable Pseudorandom Number Generators (and Almost as Fast) [video]
This video presents LXM, a new family of pseudorandom number generators that are both splittable and fast, offering improved performance and statistical quality compared to existing splittable PRNGs.
Background
- Pseudorandom number generators (PRNGs) produce sequences that appear random, used in simulations, cryptography, and machine learning. "Splittable" PRNGs can be forked into independent child generators — essential for parallel computing where many threads each need their own random stream.
- LXM (LCG + XOR + Multiply) is a family of splittable PRNGs proposed in 2021 by Daniel Lemire and Melissa O'Neill, designed to be both fast and statistically robust.
- This talk presents improvements making LXM faster while fixing weaknesses found in earlier splittable generators like Java's SplittableRandom and the Threefry/Philox family.
- Why it matters: Poor random number generation in parallel code can silently corrupt simulation results or degrade performance. Better splittable PRNGs directly improve the reliability and speed of scientific computing and AI training.