HPC 101

page:

HPC 101 for Faculty

A practical introduction for researchers, instructors, and lab leaders.

1. What Is HPC?

High-performance computing (HPC) uses parallel computing, clusters, GPUs, and large-scale storage to solve problems too large or slow for a desktop system.

Think of HPC as:

  • A large shared supercomputer made of many nodes
  • Designed for batch, parallel, and accelerated workloads
  • Accessible via remote login (SSH, web portals, notebooks)

2. Why Faculty Should Care

Research

  • Run simulations faster (CFD, FEA, molecular modeling)
  • Train ML/AI models at scale
  • Analyze large datasets (bioinformatics, astronomy, climate science)
  • Enable research that would be impossible locally

Teaching

  • Integrate parallel and high-performance concepts into courses
  • Provide students access to real computing resources
  • Use HPC for capstones, labs, or data-heavy coursework

Funding & Proposals

  • HPC strengthens grant applications (NIH, NSF, DoD)
  • Supports required data/computing plans

3. HPC Architecture: The Essentials

Cluster Components

  • Login/Head Node: where users connect, manage files, and submit jobs
  • Compute Nodes: where jobs actually run
  • Scheduler/Queue: manages resources (SLURM, PBS, LSF)

Hardware

  • CPU Nodes: many cores for parallel workloads
  • GPU Nodes: accelerators for ML and simulations
  • High-Speed Interconnect: Infiniband, Omnipath
  • Parallel Storage: Lustre, GPFS, BeeGFS

Software

Environment modules provide compilers, libraries, MPI, CUDA, Python stacks, etc.

4. How HPC Works (Operationally)

  1. Connect to the cluster (SSH, web portal, JupyterLab)
  2. Prepare a job script:
    1. CPUs/GPUs
    2. memory
    3. time limit
    4. partition/queue
  3. Submit with `sbatch`
  4. Monitor with `squeue` or `sacct`
  5. Collect output from work or scratch directories

5. Typical Research Workflows

Data Science / Machine Learning

  • Upload data
  • Activate conda or module environment
  • Submit GPU training job
  • Analyze output results

Simulation (CFD, physics, materials)

  • Load solver module
  • Prepare input deck
  • Run MPI job
  • Post-process results

Bioinformatics

  • Use tools such as BLAST, GATK, BWA
  • Submit pipeline jobs
  • Retrieve FASTA/BAM outputs

6. Best Practices for Faculty

For Research Groups

  • Train students on the scheduler (e.g., SLURM)
  • Use Git or other version control
  • Use shared directories and consistent data management
  • Prefer conda or virtual environments to avoid dependency conflicts

For Course Integration

  • Provide lightweight onboarding:
    • Linux basics
    • job scheduling
    • JupyterLab
    • quotas and usage policies
  • Use simple class assignments:
    • parallel Monte Carlo
    • GPU ML training
    • embarrassingly parallel data processing

For Proposal Writing

  • Cite cluster capabilities (nodes, GPUs, storage, CPU/GPU hours)
  • Emphasize scalability and reproducibility
  • Reference campus or national centers when applicable

7. Common Pain Points and Fixes

Pain Point || Cause || Solution
Job stuck in queue || Requested too many resources || Reduce CPUs/GPUs or use correct partition
Module not found || Wrong environment || Load correct module or use conda environment
Out-of-memory crash || Underestimated memory || Increase `--mem` or reduce processes
Python works locally but not on cluster || Missing dependencies || Use reproducible conda env
Students struggle with Linux || Steep learning curve || Use JupyterLab for teaching

8. HPC Etiquette

  • Do not run heavy jobs on the login node
  • Avoid storing large raw datasets in home directories
  • Follow fair-use and policy guidelines
  • Cite the HPC center in publications

9. Resources to Share with Students

  • SLURM, Linux, and module cheat sheets
  • Sample job scripts (CPU, GPU, MPI, array jobs)
  • Conda environment templates
  • Discipline-specific workflow examples

10. Optional: Materials I Can Generate

  • "HPC 101 for Faculty" slide deck
  • 2-page printable cheat sheet
  • HPC-ready course labs
  • SLURM job script templates
  • One-page grant-ready HPC summary
  • Custom version tailored to your institution's cluster

If you'd like, I can also produce: 

    a PmWiki sidebar version

    a PmWiki category page structure

    a shorter 1-page PmWiki version suitable for an internal documentation portal

Just tell me!

---

HPC Resources

Quick Start

For Faculty

Tools & Services

Help & Support

Page last modified on November 17, 2025, at 03:43 PM EST

|--Created: Sergey A. Uzunyan -|- Site Config --|