Dec 02 2024

Attds: Carl, Evan, Joseph, Gabriel, Nicole, Patrick, Stephen

Note-taker

General Topics

• NCCL Backend
  • Some question of how to handle initialization
    • Test system with two GPUs
    • Carl will give it a try - Jan may know SON systems with NCCL
• Sandia CI
  • AT2 is working for external contributions, Carl will stand up a test
• Supercomputing
  • Stephen + Cedric + Julien + Tony high-level meeting
  • "KaMPIng: Flexible and (Near) Zero-Overhead C++ Bindings for MPI"
    • Stephen, Evan, Riley attended, Tony talked to the authors
  • OpenMPI BoF
    • MPI ABI definition in progress
  • MPI Standards BoF
    • Patrick & Vivek & Nicole
    • Updates from committees ( ABI, fault tolerance, stream triggering discussed ), limited consensus.
  • ExaMPI Workshop
    • Kokkos + Cabana paper, may be a nice benchmark for us
    • Panel discussion about the future of MPI: standard may follow implementations if they do something good

Round Table

• Carl
  • NCCL testing & CI
• Gabriel
  • Cedric will also be testing NCCL
• Vivek
  • SC discussions
• Nicole
  • Ideas on persistence for Kokkos Comm, email coming