Jean Zay: Eviden BullSequana XH3000, HPE SGI 8600 computer

• Work in progress Jean Zay H100 extension

• Jean Zay users migration plan to other national centers after 05/02/2024 decommissioning
• Extension of the Jean Zay supercomputer to 28 Pflop/s

General information

• Jean Zay introduction
• Hardware and software configurations
• Requesting computing hours at IDRIS on Jean Zay
• The disk spaces
• SSH connections with certificates (instead of SSH keys)
• SSH public key fingerprints of the IDRIS servers
• Access while teleworking or on mission
• Recommendations for acknowledgements

Computing environment

• Passwords management
• Access and shells
• The module command
• Computing hours accounting
• Regulating of computing hours
• Guide for the multi-project user
• Transitioning between two projects
• Access control list (ACL) permissions
• Containers : Usage of Singularity
• Database management

CPU Compilation

• The Fortran and C/C++ compilation system (Intel)
• Principal options of the Intel Fortran and C/C++ compilers
• MPI parallel code
• OpenMP parallel code
• Calling C functions from Fortran
• Calling Fortran procedures from C
• CMake usage

GPU Compilation

• PGI/NVIDIA compilation system
• OpenACC parallel code
• Multi-GPU Cuda-aware MPI and GPUDirect in batch

Execution/commands of a CPU code

• CPU Slurm partitions
• Slurm commands
• Interactive execution
• Sequential job in batch
• OpenMP parallel code in batch
• MPI parallel code in batch
• Hybrid MPI/OpenMP code in batch
• MPMD execution of a code coupling in batch
• Multi-step and cascade jobs
• Collection of similar jobs: Job Array
• Memory allocation on CPU partitions
• SSH connection to a compute node

Execution/commands of a GPU code

• GPU Slurm partitions
• Slurm commands
• Interactive execution
• Single-GPU code in a batch job
• Multi-GPU code with MPI in a batch job
• Multi-GPU MPI CUDA-aware code and GPUDirect in batch
• Usage of CUDA MPS
• Multi-step and cascade jobs
• Collection of similar jobs: Job Array
• Memory allocation on GPU partitions
• SSH connection to a compute node

Debugging

• Compiler debugging options
• DDT: ARM advanced debugging tool

Performance analysis and optimisation

• Time measurement tools
• Arm quick profiling tool : Performance Reports
• Intel quick profiling tool : APS
• MAP: Arm advanced profiling tool
• VTune: Intel advanced profiling tool
• ITAC: Intel Trace Analyzer Collector
• SCALASCA: MPI and OpenMP profiling tool
• MUST : detection of usage errors of MPI
• idrmem: Memory measurement tools
• Nsight Systems: NVIDIA profiling tool
• TensorFlow and PyTorch profiling tools

CPU software and libraries

• Python
• Scientific libraries
• Atomistic simulation software

GPU and AI software and libraries

• Scientific libraries
• Atomistic simulation software
• Software for artificial intelligence
• Personal Python environment
• Access to JupyterHub (JupyterLab, TensorBoard, ...)
• In-depth documentation for AI research

Pre-/post-processing tools

• The pre/post-processing, visu, and archive Slurm partitions
• Libraries and input/output tools (HDF5, NCO, ...)
• IPSL: THREDDS at IDRIS

Visualisation tools

• Visualisation nodes
• HPC visualisation tools
• AI visualisation tools