-
Notifications
You must be signed in to change notification settings - Fork 49
Benchmark Neighbor performance
Sam Reeve edited this page Oct 16, 2023
·
4 revisions
The plot below shows performance of the VerletList
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 creation of the neighbor list (searching for neighbors in construction of the VerletList
object). Iterate
refers to looping over all particle neighbors (in an extremely simple summation kernel). Each line represents a single neighbor cutoff distance (where the larger cutoffs result in more total work and more time).
Default parameters with the commandline "large" setting were used for these results.
Cabana - A Co-Designed Library for Exascale Particle Simulations