Skip to content

2021.02.10 Meeting Notes

Andrew Gaspar edited this page Feb 10, 2021 · 2 revisions

Agenda

  • Individual/group updates
  • Scaling run results
  • GPU Hackathon (@pgrete)
  • Review non-WIP PRs

Individual/Group Updates

LANL CS

Andrew handed off sparse block work to Jonas.

Jonas is prioritizing sparse work. He has been fighting with Cray. Proxy app is stalled a little bit.

Joshua is working on performance CI. Spectrum MPI fixed issues with Darwin CI.

Carola got RIOT building on RZAnsel.

LANL RIOT

Jonah is debugging scale test with Carola. Jonah is doing KH simulation, and Carola is doing blast wave.

Joshua Brown will help Jonah/Carola use machine config file.

Ben needs code review on https://github.com/lanl/parthenon/pull/404

Joshua Dolence needs some input on some physics code - fluxes on faces in 1D.

Joshua Dolence says face centered field data is needed.

AthenaPK

Phil will add documentation for enrollment of refinement conditions and problem generations after command line and input file have been parsed. He spent more time than planned getting scaling tests running.

Kyle works at Argonne. He'd like to build an ECP proxy app based on KAthena.

Scaling run results

Phil would like to do performance runs on 12 Voltas - we could run that on Darwin on three 4-GPU nodes.

Phil notices that load balancing can take three to four minutes per-cycle.

Galen recommends using a pool allocator like Umpire.

Load Balancing could benefit from pack in one.

Non AMR performance is satisfying.

Let's try not to stress AMR for performance runs.

Galen did submit scaling run. 1 up to 2048 nodes on Sierra - 1 rank per GPU. Sitting in queue. Has not queued MPS runs yet.

Jonah thinks he needs to re-think their problem. Basically needs to translate blast wave problem. Need collaboration with AthenaPK people to compare code scaling.

GPU Hackathon

Argonne GPU Hackathon members:

  • Phil
  • Forrest
  • Jonas
  • Jonah
  • Kyle (interested)

Phil will sign up.

Clone this wiki locally