Dear NERSC Users, Allocation Year 2024 has begun! When Perlmutter returns from maintenance today, jobs will be run against AY2024 projects (though charging will not begin until after midnight Pacific time, on January 18). Any jobs in the queue as of 6 am (Pacific time) this morning that are not associated with a continuing user in a continuing project will be deleted during today's Perlmutter maintenance. Any user account not associated with a continuing or new project will begin the process of discontinuation. From now until March 16, 2024, these accounts will be able to log into NERSC resources (including Perlmutter), but will not be able to submit jobs. If you find that *you are unable to submit jobs on Perlmutter*, *please check in Iris* whether you are a member of an AY 2024 project, and if not, *contact your PI to request to be added back*. (Note that Iris is down this morning through 9 am, to update its data for the new allocation year.) An additional reminder that during this maintenance, the default *Cray Programming Environment (CPE) will be updated to 23.12 <https://github.com/PE-Cray/cpe-changelog/blob/main/ex/cpe-23.12-sles15-sp5-FullReleaseNotes.txt>*. As a consequence of this update: - *The default cudatoolkit module version will be updated to cudatoolkit/12.2.* The new version of Cray MPICH in the new CPE requires CUDA 12 for GPU-aware MPI. - *The gcc compiler will now be provided by the underlying SLES OS.* This means that the native system gcc at /usr/bin/gcc will be updated to version 12.3 and PrgEnv-gnu will now offer a gcc-native module instead of a gcc module. *We expect most users should not need to re-compile their code,* though we recommend testing existing applications in the new environment. - *For applications that require CUDA 11* or are incompatible in some other way with the new default environment, module load cpe/23.03 will restore most of the prior environment. - *Conda environments with packages from the Anaconda defaults/main channel and locally built packages that are linked with libraries in the CPE may experience compatibility issues with gcc libraries. *We recommend using the conda-forge channel which provides a more recent gcc library that is compatible with the new CPE. The conda and python modules will be updated to include the conda-forge channel by default. - *The nvcc compiler provided by cudatoolkit 11.X modules is incompatible with the updated system gcc, which may be a problem in non-GNU programming environments such as PrgEnv-nvidia.* In such cases, it may be helpful to load the gcc-mixed/11.2.0 module to provide a compatible gcc for nvcc to use other than the system gcc. We expect this will impact only a handful of users. Thanks for being a NERSC user, and we look forward to working with you in Allocation Year 2024! Regards, -Rebecca -- Rebecca Hartman-Baker, Ph.D User Engagement Group Leader National Energy Research Scientific Computing Center | Berkeley Lab rjhartmanbaker@lbl.gov | phone: (510) 486-4810 fax: (510) 486-6459 Pronouns: she/her/hers _______________________________________________ Users mailing list Users@nersc.gov