-
Notifications
You must be signed in to change notification settings - Fork 50
Benchmark Sorting performance
Sam Reeve edited this page Oct 13, 2023
·
4 revisions
The plots below show performance of sorting (top) and binning (bottom) of particles on the ORNL Frontier supercomputer. Both CPU and GPU performance are compared as a function of total particle count (with a single MPI rank).
Create
refers to building the permutation vector (construction of the binning object), while sort
refers to actually sorting (or binning) the particles. Each line represents a single hardware and binning density (where sort uses a bin size of ones), as well as sorting a single property (slice) or all particle properties (AoSoA).
Default parameters with the commandline "large" setting were used for these results.
Cabana - A Co-Designed Library for Exascale Particle Simulations