Benchmark Particle Grid Interpolation performance
The plots below show performance of particle-grid interpolation on the ORNL Frontier supercomputer. Both CPU and GPU performance are compared as a function of total grid points per rank (with eight MPI ranks).
The labels scalar, vector, and tensor denote different types of interpolation. Each point represents a single number of particles per grid cell and single interpolation operation (value, gradient, divergence).
Finally, fused refers to a synthetic case where all scalar, vector, and tensor operations were combined in a single kernel to investigate latency.
Default parameters with the commandline "large" setting were used for these results.