Difference between revisions of "Teleconference 2024-01-17"

From VTKM
Jump to navigation Jump to search
(Created page with "Attendees: == ECP Updates == * Aurora available with no allocations or restrictions ** Access through April 2024 ** Email support@alcf.anl.gov to get an account on Aurora (...")
 
 
Line 1: Line 1:
Attendees:  
+
Attendees: Ken Moreland (ORNL), Mark Bolstad (SNL), Silvio Rizzi (ANL), Abhi Yenpure (Kitware), Jeff Amstutz (Nvidia), Jay Wang (ORNL), Dave Pugmire (ORNL), Tushar Athawale (ORNL), Nicole Marsaglia (LLNL)
  
 
== ECP Updates ==
 
== ECP Updates ==
Line 14: Line 14:
 
== Roundtable ==
 
== Roundtable ==
  
SNL is looking VTK-m with Kokkos contention on a GPU when the simulation is also using Kokkos. First, is there any contention and if so how bad is it?
+
Silvio was able to get the compile of VTK-m using Ascent types to work on Aurora, but only by going to a compiler released in May. Tracing this down is ongoing.
  
Abhi having trouble compiling VTK-m under VTK on Aurora. Problems like linking pop up. Also problems about how much memory it requires to compile a filter like Gradient.
+
Dave is optimizing the Poincaré plot code by tuning thread parameters. So far he has got speedups of about 1.8x. This is good, but does not seem quite as good as he got on Nvidia.
 
 
Nicole is planning on looking at material reconstruction in VTK-m.
 
 
 
There is some issue with the tube filter in WarpX data.
 
 
 
Dave is having some issue running the Poincaré application on Frontier. The streamlines don't seem to be running correctly on the HIP hardware.
 

Latest revision as of 13:22, 17 January 2024

Attendees: Ken Moreland (ORNL), Mark Bolstad (SNL), Silvio Rizzi (ANL), Abhi Yenpure (Kitware), Jeff Amstutz (Nvidia), Jay Wang (ORNL), Dave Pugmire (ORNL), Tushar Athawale (ORNL), Nicole Marsaglia (LLNL)

ECP Updates

  • Aurora available with no allocations or restrictions
    • Access through April 2024
    • Email support@alcf.anl.gov to get an account on Aurora (if you already have one on Sunspot)
  • Confluence and JIRA are gone
  • Zoom going away May 31
    • Ken will look into replacing with another account
  • IJHPCA paper
  • High Performance Software Foundation

Roundtable

Silvio was able to get the compile of VTK-m using Ascent types to work on Aurora, but only by going to a compiler released in May. Tracing this down is ongoing.

Dave is optimizing the Poincaré plot code by tuning thread parameters. So far he has got speedups of about 1.8x. This is good, but does not seem quite as good as he got on Nvidia.