Скачать или смотреть Demonstrating libEnsemble on the Frontier Exascale supercomputer (with GPU application)

Demonstrating libEnsemble on the Frontier Exascale supercomputer (with GPU application)

This video gives a small-scale demonstration of using libEnsemble to run multiple GPU applications on a node on the Frontier supercomputer at Oakridge National Laboratory. Dynamic resource allocation is also demonstrated.

libEnsemble is a Python tool for running dynamic ensembles, developed under the ECP project "Preparing PETSc/TAO for Exascale".

libEnsemble can be found at: https://github.com/Libensemble/libensemble

The example codes in the video can be found at the following locations.

Forces GPU
Each simulation is given a number of particles randomly chosen within a range and runs with one processor and one GPU by default.
https://github.com/Libensemble/libensemble...

Forces GPU with variable size simulations
Each simulation is run with a number of MPI ranks (and GPUs) based on the number of particles.
https://github.com/Libensemble/libensemble...

Also, see forces_multi_app for an example that combines a CPU-only application and a GPU application, which are randomly selected.
https://github.com/Libensemble/libensemble...

Video presented by: Steve Hudson