Skip to content

Oct 7 2024

Carl Pearson edited this page Oct 7, 2024 · 3 revisions

Attds: Carl, Gabriel, Cedric, Evan, Patrick, Stephen, Nicole, Vivek

Note-taker

General Topics

  • #125 Stream-Ordered wait
  • FYI, Kokkos developer meeting 11/12-11/15 @ ORNL (not KUG) (Cedric & Carl)
  • FYI Supercomputing (11/16-11/23) (Cedric, Carl, Stephen, Patrick, Evan, Nicole)
    • maybe meet up at UNM booth
  • EuroMPI happened (Hugo)
    • People seemed to understand what and why, are curious
    • Want to have some follow up with C++ / MPI people
  • Cluster happened (Nicole & Evan)
    • People like the direct view support
    • Dr. Estela Suarez (?), modular supercomputing architecture & performance portability
    • General interest, esp in GPUs

Working Groups

  • NCCL
  • CI for GPUs (Carl)
  • Applications
  • Modern C++ / MPI
  • Accelerator-Initiated Communication / Support
  • Smart NICs

Round Table

  • Carl
    • Sandia has implemented infrastructure changes, need to try them out.
  • Gabriel
    • CMake PR merged
  • Cedric
    • 2D matrix product example coming soon
  • Evan
    • Stream triggering w/Patrick
  • Patrick
    • Stream-triggered halo abstraction in Cabana, stream-triggered planning, coordination with HPE on an implementation (Tioga and Chicoma)
    • Using fine-grained zero copy memory on MI300/MI250 to remove an explicit copy
    • On MI250 you can't use device atomics on fine-grained memory
    • Do users need to be able to query memory properties to do the right thing?
  • Nicole
    • Working with Cabana on Tioga
  • Vivek
    • No updates, still thinking about Kokkos Tools + MPI
    • Useful for our own understanding of Kokkos Comm