Skip to content

Cluster Basics

New to NCShare? This page explains how the cluster is put together, how your work gets run, and the terms you'll see throughout these guides.

What is NCShare?

NCShare Compute as a Service (CaaS) and GPU as a Service (GaaS) is a program funded by the U.S. National Science Foundation that gives researchers and educators at participating institutions computing and storage at no cost. It supports everything from interactive sessions with Jupyter and RStudio to parallel workflows and GPU-accelerated work on NVIDIA H200s.

How work runs on NCShare

  1. Your computer

    Connect with SSH, VS Code, or OnDemand in a browser.

  2. Login node

    Edit files, install software, and submit jobs. Not for heavy computation.

  3. Slurm scheduler

    Queues your job and picks nodes in the partition you asked for.

  4. Compute nodes

    Run your job on CPU nodes or H200 GPU nodes with the resources Slurm allocated.

Shared storage (home and work) is available on every node, so your files are where your job needs them.

Choosing how to work

  • Open OnDemand

    Jupyter Lab, RStudio, a desktop, and a shell in your web browser, running on cluster resources.

    OnDemand

  • Command line

    Connect with SSH or VS Code and work directly with Slurm, for full control over your jobs.

    Login

  • GPUs

    NVIDIA H200 GPUs through Slurm or OnDemand, available on request through your institution.

    GPUs

  • Courses

    Semester-long Jupyter and RStudio containers in Container Manager, built for classrooms.

    Courses

Terminology

The hardware

node
A physical machine in the cluster. NCShare has a shared login node, CPU compute nodes, and H200 GPU nodes. See Architecture for the full specifications.
CPU
On NCShare, a CPU in Slurm is a hardware thread. Each physical core provides two.
GPU
A graphics processing unit. NCShare's NVIDIA H200 GPUs accelerate deep learning, simulations, and other highly parallel scientific computing. See GPUs.

Running work

Slurm
The open-source workload manager that schedules and runs jobs on NCShare. It starts your job at the soonest time the requested resources are free, so jobs that ask for only what they need tend to start sooner.
job
A task or set of tasks submitted to Slurm to run on the cluster.
batch job
Work that runs without you watching it. You write the commands into a batch script and submit it with sbatch. See Scheduling Jobs.
interactive session
A real-time session on a compute node, for coding, debugging, or watching work as it runs. Start one through OnDemand or with srun.
partition
A group of compute nodes with its own limits, such as who can use it and how long jobs can run. See Partitions.
container
A packaged software environment. NCShare runs Apptainer containers on the cluster, and Container Manager containers for courses.

Where to go next