Conference call notes 20200415

(back to Conference calls)

Notes on the 145th EasyBuild conference call, Wednesday Apr 15th 2020 (8:00 UTC - 10:00 CEST)

Attendees

Alphabetical list of attendees (7):

• Miguel Dias Costa (Univ. of Singapore)
• Pablo Escobar (sciCORE, University of Basel)
• Victor Rusu (CSCS, Switzerland)
• Kenneth Hoste (HPC-UGent, Belgium)
• Adam Huffman (Big Data Institute, Oxford, UK)
• Åke Sandgren (Umeå University, Sweden)
• Alexandre Strube (JSC, Germany)

Agenda

• overview of recent changes included in EasyBuild v4.2.0
• update on 2020a update of common toolchains
• Q&A

Overview of recent changes included in EasyBuild v4.2.0

• released yesterday (April 14th)
• highlights: https://github.com/easybuilders/easybuild/releases/tag/easybuild-v4.2.0
• small hick-up in release of easybuild package on PyPI, quickly resolved
  • all details in https://github.com/easybuilders/easybuild/pull/615#issuecomment-613540406

Update on 2020a common toolchains

• see project dashboard: , and see also the 2020a branch and milestone
• foss/2020a
  • GCC 9.3.0 + binutils 2.34
  • OpenMPI 4.0.3
  • OpenBLAS 0.3.9
  • ScaLAPACK 2.1.0
  • FFTW 3.3.8
• intel/2020a
  • based on GCC 9.3.0 + binutils 2.34
  • planning to update to iccifort/imkl 2020 update 1 + impi 2019 update 7, see https://github.com/easybuilders/easybuild-easyconfigs/pull/10295
  • iccifort 2020.1.217 (a.k.a. 2020 update 1, v19.1.1)
  • impi 2019.7.217
    • some concerns here (with both update 6 & 7)
    • see https://github.com/easybuilders/easybuild-easyconfigs/issues/10314
    • we need to sort this out ASAP
  • imkl 2020.1.217
• not compatible CUDA yet, so fosscuda/2020a on hold
• TODO:
  • figure out what to do with impi 2019 update 6/7
    • see https://github.com/easybuilders/easybuild-easyconfigs/issues/10314
  • more testing with (updated) intel/2020a
  • collapse 2020a branch into develop
    • gradually: first GCCcore-9.3.0 easyconfigs, then foss/2020a
    • ensure PRs to develop branch can be re-tested (so not too many easyconfigs in a single PR)
  • potential way forward is to only collapse foss/2020a into develop, put intel/2020a on hold...
  • figure out how to get rid of 2020a branch, what to do with open PRs to 2020a?

Q&A

• Pablo: GCC + OpenMPI GPU offloading
  • anyone else playing with this?
  • requires significant changes in GCC easyblock
  • Victor: ongoing ECP project on this to get this to work well
    • requires recent version of OpenACC
    • still some bugs
  • any docs on this?
    • only GCC docs, which are not easy to digest...