Skip to content

FF chevron

Why use the HPC cluster?

Let's determine whether to:

  • Use your own laptop
  • Buy your own server
  • Use the university's HPC cluster

What is a Supercomputer?

A Supercomputer is not a single machine; instead, it's a cluster of server-class machines (nodes) inter-connected with high-speed networking. Each of those computer nodes contains many CPU cores and a large amount of memory (RAM). Some nodes also have high-end Graphics Processing Units (GPUs). In addition, an HPC cluster also has a very large disk storage system that uses a high-performance parallel filesystem. See the specs for UMD's Zaratan HPC cluster.

How are supercomputers or HPC clusters useful?

  • No need to purchase and house your own computer hardware
  • Can run thousands of instances of similar calculations in parallel
  • Able to run large parallel jobs that require large amounts of CPU cores and memory
  • Often run long jobs for multiple days
  • An HPC cluster is a time-share system--it is nearly 100% used nearly 100% of the time, reducing the waste of idle compute cycles and power
  • High-end GPUs are available for modern AI and Machine Learning tasks
  • Software needed for your research may already be installed on the cluster
  • Professionally maintained by a group of HPC engineers

Shall I use my Laptop/Desktop to run compute tasks?

Yes, certainly, you should use your laptop or desktop to:

  • Run single-threaded R/Julia/Python/etc jobs that finish in a few minutes
  • Use RStudio/Jupyter Notebook/Matlab Desktop/etc for code development or trial-and-error testing
  • Conduct trial runs to understand how to scale out or paralellize your tasks

Shall I buy my own server so I can use it anytime I want?

If you purchase your own server-class machine, please consider:

  • Where will you house it, and are there recurring costs?
  • Who will manage the hardware service calls?
  • Who will maintain the Operating System, patches, user accounts, etc?
  • Who will install the research software you need?
  • What happens when your needs outgrow the server's specs?
  • What happens when the hardware warranty runs out?
  • The HPC cluster is already built, and available now. Get an account and adapt your research workflow to make use of the HPC cluster.

How to contribute to the UMD HPC environment?

  • With previous clusters, we used to offer co-op or condo model of adding hardware which were purchased by researchers/groups. We no longer support this model, since:

    • The co-op hardware are generally not 100% available to other members of the HPC user community even when they are idle.
    • We end up with too many different processor micro-architectures which prohibits us from properly optimizing the software packages.

  • If you wish to purchase HPC resources:

    • Smaller amounts of Service Units (SUs) and disk space are available for purchase directly.
    • If your Department or College would like to contribute a significant amount towards the expansion of the HPC cluster, please contact us and we will work with you.

  • Another big contribution you can make towards advancing HPC usage at UMD is to acknowledge the UMD HPC resources with a citation in your publication.