Guide for using containers on Isambard AI (BriCS) supercomputers with Podman-HPC and Apptainer (formerly Singularity). Use this skill whenever a user asks about running containers on Isambard, using podman-hpc or apptainer/singularity, pulling or building container images, running containers on compute nodes or across multiple nodes with MPI or NCCL, GPU access inside containers, squashfs image migration, the brics/nccl or brics/apptainer-multi-node modules, using NGC (Nvidia GPU Cloud) containers on Isambard, or getting full Slingshot 11 interconnect bandwidth from containerised workloads. Also trigger for questions about aarch64-compatible container images, --fakeroot builds, the /host/adapt.sh entrypoint, or any container-related workflow on an HPE Cray system.
Guide for using GPUs and CUDA on Isambard AI (BriCS) supercomputers (Isambard-AI Phase 1 and Phase 2, which have NVIDIA GH200 Grace Hopper GPUs). Use this skill whenever a user asks about CUDA on Isambard, loading the cudatoolkit or nvhpc modules, compiling CUDA kernels with nvcc, the correct -arch flag for the GH200 (sm_90), CUDA forward compatibility, using a newer CUDA version than the installed driver supports, the NVIDIA HPC SDK on Isambard, NGC container images for GPU workloads, nvidia-smi on compute nodes, or diagnosing CUDA version mismatches. Also trigger for general GPU programming setup questions on Isambard, even if the user doesn't say "CUDA" explicitly. Note: GPUs are only on Isambard-AI (Phase 1 and Phase 2) — not Isambard 3.
Guidance for writing, reviewing, and running AI-generated code responsibly on Bristol Centre for Supercomputing (BriCS) shared HPC resources: Isambard-AI and Isambard 3. Use this skill whenever generating, adapting, or debugging code intended to run on BriCS facilities, writing Slurm job scripts, managing storage, installing software, or ensuring compliance with BriCS policies.
Guide for using the modules system, programming environments, compilers, and profiling tools on Isambard AI (BriCS) supercomputers (Isambard-AI and Isambard 3). Use this skill whenever a user asks about loading modules on Isambard, the Cray Programming Environment (PrgEnv-gnu, PrgEnv-cray), compiler wrappers (cc, CC, ftn), GCC or NVIDIA compilers, compiler flags for Arm/Grace (neoverse-v2), BriCS-supplied modules (brics/nccl, brics/tmux, etc.), profiling with gprof, Cray perftools, or Nvidia Nsight Systems. Also trigger for questions about linking MPI or scientific libraries on Isambard, or troubleshooting compiler and build issues on the Cray HPE system — even if the user doesn't explicitly say "modules".
Guide for using MPI on Isambard AI (BriCS) supercomputers (Isambard-AI and Isambard 3). Use this skill whenever a user asks about MPI on Isambard, running multi-node MPI jobs, Cray MPICH, cray-mpich, PrgEnv-gnu or PrgEnv-cray for MPI, the PMI or PMIx process manager interface, srun --mpi flags (cray_shasta, pmi2, pmix), installing OpenMPI or MPICH with conda or from source, mpi4py on Isambard, libfabric and Slingshot 11 MPI performance, or why mpirun and mpiexec should not be used on Isambard. Also trigger for questions about linking MPI libraries, compiler wrappers with MPI (mpicc, mpicxx, mpif90), or any multi-node communication setup on an HPE Cray system.
Guide for using NCCL (NVIDIA Collective Communications Library) on Isambard AI (BriCS) supercomputers. Use this skill whenever a user asks about NCCL on Isambard, multi-node GPU communication, the aws-ofi-nccl plugin, Slingshot 11 RDMA for GPU collectives, the brics/nccl module, NCCL environment variables (NCCL_NET, NCCL_SOCKET_IFNAME, FI_CXI_* variables), building NCCL or aws-ofi-nccl from source, benchmarking NCCL with nccl-tests, NCCL in containers (Apptainer/Singularity), or diagnosing slow multi-node GPU bandwidth on Isambard-AI. Also trigger for questions about GPUDirect RDMA on Slingshot, distributed training performance, or getting full interconnect bandwidth for GPU collective operations — even if the user doesn't explicitly say "NCCL". Note: NCCL is only supported on Isambard-AI Phase 1 and Phase 2 — not Isambard 3.
Guide for installing and managing Python environments on Isambard AI (BriCS) supercomputers. Use this skill whenever a user asks about Python on Isambard, setting up Conda/Miniforge, using uv, installing pip packages, building from source on aarch64/Arm64, mpi4py on HPC, virtual environments on Isambard, or troubleshooting Python package builds for Linux Arm64. Also trigger for questions about Cray Python, build isolation, pyproject.toml on Arm, or finding aarch64-compatible packages — even if the user doesn't say "Isambard" explicitly but is clearly working on an HPC or Arm Linux environment.
Guide for submitting, monitoring, and managing HPC jobs using the Slurm workload manager on Isambard-AI (GH200 GPU nodes) and Isambard 3 (Grace CPU and MACS nodes). Use this skill whenever a user asks about Slurm on Isambard, writing sbatch scripts, requesting GPUs or CPU resources, srun or salloc, job arrays, job dependencies, multi-node jobs, hybrid MPI/OpenMP jobs, QOS limits, scheduler flexibility (--time-min, --nodes range), --exclusive, sacct, polling intervals, job accounting, or why a job is stuck in PENDING, failing, or hitting resource limits. Also trigger for questions about acceptable use of the Slurm queue, what PENDING reasons mean, how to chain jobs, or how to debug a running job — even if the user doesn't explicitly say "Slurm".