Skip to content
Science

November 13, 2024

November 13, 2024

Present

Geoffrey Fox, Gregor von Laszewski, Juri Papay, Gary Mazzaferro, Wes Brewer, Marisa Ahmad, Armstrong Foundjem, Piotr Luszczek, Victor Lu, Andy Cheng, Shirley Moore, Claus Weiland, Gregg Barrett, Satoshi Iwata, Piero Coronica, Azza Ahmad, Steve Farrell, Murali Emani, Sharma Lee, Christine Kirkpatrick, Gavin Mitchell Farrell

Tentative Agenda

New Members

White Papers

  • Christine will get out the white paper on energy which has been restored to roughly original form and sent to Gregor
  • We will spend time, up to 20 minutes, on AI Readiness paper at next meeting which is the day before Thanksgiving
  • Christine noted relevance of open models and data in AI readiness
  • We will also spend time, up to 20 minutes, on benchmark Carpentry paper at next meeting
  • Shirley Moore will add material on benchmark tools
  • Add a survey of benchmark projects
  • Authors should add their names and material

Other Discussion

  • The plan for this meeting was changed at last moment as the plan to merge science and HPC working groups was postponed for consideration of other futures for HPC group. The preparation for the meeting was not properly done and Geoffrey apologized for this.
  • Juri noted that Summit is closing very soon
  • Armstrong can get access to UVA computers if necessary
  • Marisa noted the MLCommons SC Happy hour at Time: 7-9pm, (following the SC ‘24 MLPerf BoF Session from 5:15 to 6:45 pm) and Location: Atlantucky 170 Northside Dr SW Suite 96, Atlanta, GA 30313 | map
  • Google just ordered 7 small nuclear reactors but 5-10 years to come online (Christine)
  • Waste “only” has a 300 year half life as low density
  • Christine noted that some software particularly power intense and one needs to raise understanding

Benchmarks

  • Geoffrey noted goal of identifying uptodate benchmarks in areas like weather with many recent deep learning innovations. Gary noted NOAA’s initiative https://sab.noaa.gov/wp-content/uploads/4.0-DL4NWP_NOAAResponse_Nov2024.pdf here and Shirley’s comments are spelled separately later.
  • Gavin noted his experience with bioinformatics benchmarks
  • The activity in Materials Science Foundation models was noted
  • DOE and AI Alliance active
  • Improve benchmarks so easier to analyze with complete ontologies and no free text
  • Energy analysis should be part of benchmarks
  • Gregor and WEs noted Mom6 benchmark delayed by HPE but should complete soon

Comments from Gary Mazzaferro

Comments from Shirley Moore