General move towards FlexiBLAS with MKL, BLIS and OpenBLAS as backends.
Applies to gcccoreflexiblas (as opposed to gcccoremkl), and the upper level toolchains, also with other compilers.
Version changes in packages of major importance
GCC updated to 12.3.0
ParaStationMPI updated to 5.9.0-1
OpenMPI updated to 4.1.5
Software distribution-related changes
Python is now just a base installation with basic extensions. Most of the other Python extensions are part now of the Python-bundle-PyPI/2023.06 module.
Perl is now just a base installation with basic extensions. Most of the other Perl extensions are part now of the Perl-bundle-CPAN/5.36.1 module.
WIP packages
Deprecated packages
nvidia-driver module no longer available. The functionality is now split between RPMs installed in the OS and the OpenGL module.
libelf is not used as a dependency, so it has been marked as deprecated. The library itself is part of the elfutils package.
LCov was not used anymore
SuperLU was not used anymore
cppcheck was not used anymore
Renamed packages
Changelog for stage 2023
General software stack changes
Custom installations are now supported via the UserInstallations module (see documentation)
OpenMPI now supports UCC (which supports SHARP and NCCL as backends)
Version changes in packages of major importance
GCC updated to 11.3.0
ParaStationMPI updated to 5.6.0-1
OpenMPI updated to 4.1.4
IntelMPI updated to 2021.6.0
Software distribution-related changes
imkl has been split in imkl, which contains the binary libraries for most of the MKL functionality, and imkl-FFTW, which contains the FFTW MPI-wrappers.
IntelMPI now leverages the MPICH ABI compatibility with ParaStationMPI and reuses its stack (packages are no longer compiled for IntelMPI in a separate toolchain)
gomkl toolchain (GCC + OpenMPI + MKL) was replaced by the toolchain foss (GCC + OpenMPI + FlexiBLAS). FlexiBLAS allows you to change between MKL, OpenBLAS and BLIS. You can list all installed backends and the complete configuration: flexiblas list. Changing the backend can be done for example by seeting FLEXIBLAS=OPENBLAS. C.f. https://gitlab.mpi-magdeburg.mpg.de/software/flexiblas-release#usage-example
WIP packages
Deprecated packages
pkg-config module no longer available. Replaced with pkgconf
Subversion and Mercurial are no longer supported in our stack.
Grace is no longer supported in our stack.
iomkl toolchain (intel-compilers, OpenMPI, MKL) was deprecated
Renamed packages
Changelog for stage 2022
First commit - 18.03.2022
General software stack changes
The last 3 production stages have been made visible. The oldest one is deprecated and will not be visible by default when work on a new stage begins
Older stages are visible via ml use $OTHERSTAGES
BullMPI has been added as an MPI runtime with GCC on JURECA-DC
mpi-settings has been divided in mpi-settings (for MPI-runtime specific settings) and UCX-settings.
Version changes in packages of major importance
GCC updated to 11.2.0
NVHPC updated to 22.1
intel-compilers updated to 2021.4.0
ParaStationMPI updated to 5.5.0-1
IntelMPI updated to 2021.4.0
OpenMPI updated to 4.1.2
Software distribution-related changes
SciPy-Stack is now a wrapper around other packages. SciPy-Stack/2021b includes SciPy-bundle/2021.10, matplotlib/3.4.3, Seaborn/0.11.2, sympy/1.8 and xarray/0.20.1.
kim-api Moved to GCCcore
Octave Installed in GCCcore
scikit Split in multiple packages scikit-allel, scikit-bio, scikit-build, scikit-image, scikit-learn, scikit-optimize and scikit-uplift
Boost Moved to GCCcore
Boost.Python Moved to GCCcore
WIP packages
Jupyter related packages
PyTorch
TensorFlow
Deprecated packages
intel-para module no longer exposed in the module hierarchy
basemap deprecated in favour of Cartopy
libaec
libctl deprecated due to lack of compatibility with latest guile
libfabric deprecated due to lack of usage in the current stage
meld
NCL deprecated due to lack of maintenance upstream and lack of support for GCC 11. See https://github.com/NCAR/ncl/issues/64
nvptx-tools integrated in GCC
ospcommon deprecated in favour of rkcommon
Elemental not developed anymore
HDFView latest release not compatible with latest HDF5
Julia.CUDA integrated in Julia
Julia.MPI integrated in Julia
yaff deprecated due to lack of compatibility with latest h5py at the moment