1. OIT HPC
  2. Overview

Overview of HPC Service

High Performance Computing at NC State

The Hazel High Performance Computing (HPC) cluster is a shared system that helps researchers and students run computational work that is too large, too time-consuming, or too complex for a typical desktop or lab server. HPC enables faster turnaround, larger models, and the ability to run many analyses in support of research and instruction across a broad range of disciplines.

There is no charge for using HPC compute resources.

How the HPC cluster is used

Work on the HPC cluster is submitted as a job to a scheduler. Users request the resources their work needs such as CPU cores, memory, GPUs, and run time and the scheduler runs the job when resources are available. This approach supports efficient, fair sharing while enabling jobs that scale from quick tests to large production campaigns.

Common HPC usage patterns include:

  • Large simulations that require many cores or high memory
  • Parameter sweeps and ensembles (many related runs exploring different inputs or scenarios)
  • High-throughput workloads (large numbers of independent tasks)
  • GPU-accelerated computing for workloads that benefit from GPUs (e.g., AI/ML, imaging, molecular simulation)
  • Post-processing and analysis of large outputs and datasets

What software runs on HPC

The HPC cluster supports a broad mix of software: licensed commercial applications, community-supported research codes, and custom tools developed by research groups.

Commercial applications (licensed)

Many researchers use HPC to run widely adopted commercial applications (such as ANSYS and Gaussian), running larger models, higher fidelity simulations, or more design iterations.

Community-supported research applications (open / shared)

HPC is also commonly used for open research codes maintained by scientific communities and optimized for parallel computing, such as LAMMPS, VASP, and WRF.

User-developed applications (custom code and workflows)

A substantial portion of HPC use involves research-group models, pipelines, and analysis tools written in Python, C/C++, Fortran, R, CUDA, and other languages. These applications often combine multiple steps (simulation, data processing, visualization, AI) and scale across many cores, nodes, or GPUs.

Containerized applications

Increasingly, applications are available as containers that help enhance reproducibility of results as well as ease of use. Apptainer is provided on the HPC cluster for running containers.

See the full list of available software.

Compute resources

The Hazel cluster is a heterogeneous cluster that includes state-of-the-art CPUs, GPUs, and networking while maintaining older resources as long as feasible. Currently there are:

  • On the order of 400 compute nodes with over 14,000 cores
  • Majority of nodes connected with InfiniBand
  • Several nodes with one or more attached GPUs of various models (A100, H100, H200, L40S, and others)
  • Most nodes have more than 128 GB of memory; standard configuration is now 512 GB with some 1024 GB nodes

See the Compute Resources page for full details and the Cluster Status pages for real-time availability.

Storage

  • Home directory (/home) — 1 GB per account for source code, scripts, and small executables
  • Scratch space (/share) — 20 TB per project for running applications and working with large data
  • Application storage (/usr/local/usrapps) — 100 GB per project by request, for installing larger applications and conda environments

See the Storage page for details on directory locations and size limits.

When HPC may not be the best choice

HPC is optimized for batch, compute-intensive workloads. In some situations, other computing options are a better fit:

  • Interactive, latency-sensitive work: If you need instant feedback (e.g., frequent GUI-driven interactions or real-time control), a workstation or dedicated server may be more productive.
  • Small jobs with heavy overhead: If the work finishes in seconds and needs constant re-runs, scheduler queue time may outweigh the benefits of the cluster.
  • Workloads that don't parallelize well: Software that cannot be run in parallel provides no speed advantage on a cluster.
  • Always-on services: Long-running services (web apps, databases, dashboards) are usually better hosted on a managed server platform rather than a shared batch cluster.

Getting started

We look forward to working with you

Take a look at what OIT-HPC has to offer and contact us any time to ask a question, report a problem, or schedule a consultation.