ESiWACE3 Hackathon Planning #143

Activity

1. kvrigor

   added

   tracking-issueA collection of issues.A collection of issues.

   on Oct 1, 2026
2. github-project-automation

   added this to TSMP2: Towards Stages/2026 supporton Oct 1, 2026
3. github-project-automation

   moved this to New features in TSMP2: Towards Stages/2026 supporton Oct 1, 2026
4. kvrigor

   pinned this issue on Oct 1, 2026
5. kvrigor

   added sub-issues on Oct 2, 2026

   • MPI-settings behavior on JSC systems #129
   • eCLM crashes with soil balance-check error while running EUR0275 setup — same setup runs fine under CLM5 eCLM#121
   • OASIS3-MCT fails to build under NVHPC+OpenMPI toolchain (Stages/2026) #142
   • ICON 2024.07 build fails on GCC 14.3 #141
   • Update to icon-2026.04 icon-model_coup-oas#1
6. kvrigor

   added a sub-issue on Oct 14, 2026

   • eCLM-ICON: Recent TSMP2 fixes eCLM#151
7. kvrigor

   linked a pull request that will close this issueTowards TSMP2-GPU on JUPITER #147on Oct 23, 2026
8. github-project-automation

   moved this from New features to Housekeeping in TSMP2: Towards Stages/2026 supporton Oct 23, 2026
9. kvrigor commented on Oct 23, 2026

   

   kvrigor

   on Oct 23, 2026

   Last edited by kvrigor

   MemberAuthor

   More actions

   Hackathon wrapped up yesterday. Vibe is similar to a LAN party minus the cursing and with just enough chaotic energy for everyone to keep sitting for ~9 hours straight, only remembering our human needs during lunch/coffee/bathroom breaks 🤖. More seriously, I enjoyed this hackathon setting where everybody sits on one table and focused on solving a specific set of problems--a very productive way to work! Here's what we've accomplished in the past three days:

   • Building from @kgoergen's ICONGPU 3km experiments, @s-poll were able to generate EU-3km static files for eCLM-ICONGPU and were able to run it on JUPITER. As expected eCLM-ICONGPU didn't work at first try, but Stefan was able to do patches on eCLM, ICON, and OASIS to make the setup run.
   • We could now build a fully-coupled TSMP2 model on JUPITER, with both Parflow and ICON on GPU. We didn't have time to test the full model, but this is already a good starting point for fully-coupled EU-3km runs.
   • @AGonzalezNicolas got stressed in stress-testing her JUBE scripts that are supposed to automatically benchmark the idealized test case and EU 12km. I think is a good thing! Ana revealed a lot of gaps in the TSMP2 workflow engine which need to be fixed if anybody wants to replicate our latest EU-12/EU-3 runs. We don't want to always bother Stefan about the location of the updated static files 😉.
   • @jjokella contributed fixes to successfully build eCLM-PDAF and eCLM-ParflowGPU-PDAF on JUPITER. This sets the stage for GPU-accelerated DA runs.

   For posterity I'm attaching our final presentation slides and linking to our hackathon hedgedoc notes. Thanks again Ana, Stefan, and Johannes for the good time in the hackathon!

   👍React with 👍2Reacted by Klaus GOERGEN and Daniel Caviedes-Voullième

   ❤️React with ❤️1Reacted by Ana Gonzalez-Nicolas
10. kvrigor

    closed this as completedon Oct 23, 2026
11. github-project-automation

    moved this from Housekeeping to Issues in TSMP2: Towards Stages/2026 supporton Oct 23, 2026
12. kvrigor

    removed sub-issues on Oct 23, 2026

    • MPI-settings behavior on JSC systems #129
    • eCLM crashes with soil balance-check error while running EUR0275 setup — same setup runs fine under CLM5 eCLM#121
    • OASIS3-MCT fails to build under NVHPC+OpenMPI toolchain (Stages/2026) #142
    • ICON 2024.07 build fails on GCC 14.3 #141
    • Update to icon-2026.04 icon-model_coup-oas#1
    • eCLM-ICON: Recent TSMP2 fixes eCLM#151
13. kvrigor

    removed this from TSMP2: Towards Stages/2026 supporton Oct 23, 2026
14. kvrigor

    unpinned this issue on Oct 23, 2026