Climbing the Hardware Ladder: Research Computing at HMC
This first ARCS post is a practical overview of the research computing resources open to the HMC community, and a preview of a new on-campus service coming online this summer.
If you’ve run into long runtimes, memory limits or hardware constraints on your laptop, this should help you work out what comes next.
A useful way to think about this is the Hardware Ladder (credit to Paul Nerenberg at CMC). Research computing resources span a set of rungs, from personal machines up to national supercomputers. Each step up gives you more capability, but makes the system more complex to reach and use.
Working effectively means knowing which rung fits your workload, and how to move between rungs as your needs change. ARCS can help with both.
The Hardware Ladder at a glance
| Rung | Resource | Scale and capability | Access | Friction |
|---|---|---|---|---|
| 1 | Personal machine | Local CPU or GPU, limited memory | You already have one | None |
| 2 | Project Iris, a single-node server or VM | Hundreds of GB to several TB of RAM, 100+ cores, dedicated GPUs | Direct VM lease; students can get accounts without a PI | Low |
| 3 | Hopper, the CMC consortial cluster | Multi-node, scheduled with Slurm | Account request; research use | Moderate |
| 4 | Laguna, the USC CARC regional cluster | Many nodes; browser access through OnDemand | PI-based; compute-hour credits | Moderate to high |
| 5 | ACCESS, the national supercomputers | 50+ national systems, specialized AI and ML hardware, enormous scale | Allocation application; credit-based | High |
Each rung suits a different kind of workload. Moving up usually takes more planning, coordination and familiarity with the system.
Rung 1: Your laptop
Most projects begin on a personal machine. Laptops suit development, prototyping, small datasets and early analysis: you control the environment, iteration is fast, and there’s no shared infrastructure to manage.
Signs that a workload has outgrown a laptop include long runtimes, memory limits, datasets bigger than local storage, or needing hardware such as GPUs that you don’t have. When these become routine, it’s usually time to move up a rung.
Rung 2: A single-node server or VM
Rung 2 gives you more resources without the full complexity of a shared cluster: typically more memory and CPU cores, persistent environments, and remote access over SSH.
At HMC, this layer has historically been departmental or faculty-managed systems, for example Knuth and Teapot in CS, Hyper in Math, and Gandalf and Galadriel in Chemistry. They work well, but they aren’t available to every department. Iris is designed to fill this rung for all of HMC.
Project Iris
Iris is an on-campus HPC service being deployed by ARCS. It provides virtual machines on shared, HMC-managed hardware for research and teaching.
System overview
The first node has two AMD EPYC 9965 CPUs (384 cores in total), 3 TB of RAM, 60 TB of NVMe storage and an NVIDIA RTX 4500 Blackwell GPU (32 GB). It will be housed in the McGregor Data Center, with room to add nodes over time.
A GPU-focused second node is planned but deferred because of current hardware prices. The rollout is phased: establish the service, support the first users, and expand as demand shows.
Why virtual machines?
Iris uses Proxmox to provide virtual machines rather than shared user accounts on one system. Each VM is an independent environment, which gives:
- Custom software environments. Each VM can be set up for its own research or teaching needs.
- Isolation between users. Changes in one VM don’t affect the others.
- Flexible resources. CPU, memory and storage are assigned per VM to fit the workload.
Usage models
At launch, Iris supports three uses:
- Student exploration VM. Students can request access to a shared VM without belonging to a research group: a low-friction way to learn and experiment.
- Rivendell replacement VM. A modern environment for the Chemistry workloads now on Gandalf and Galadriel, with a planned transition period.
- Faculty VM leasing. Dedicated VMs for research groups, sized to the project. Faculty can also contribute hardware or funding toward expansion.
ARCS co-administers Iris with support from central IT. A request form is coming soon; early interest is welcome.
Iris as a development environment
A VM is a controlled place to develop and test a workflow before scaling it. Once a workload’s needs are clear, it can move to larger systems such as Hopper, Laguna or ACCESS.
Rung 3: Hopper, the CMC consortial cluster
Hopper is a Slurm-managed cluster hosted at Claremont McKenna College. It’s a traditional shared cluster, with job scheduling, environment modules and support for parallel workloads.
Hopper suits workloads beyond a single machine, including multi-node jobs, large parallel processing and GPU computation. Access is currently for research use, through account requests.
Rung 4: Laguna, the USC CARC regional cluster
Laguna, run by USC’s Center for Advanced Research Computing, is a regional cluster open to institutions across Southern California, including HMC. Access is PI-based and uses compute-hour credits.
Laguna stands out for its web interface, OnDemand, which gives browser access to a shell, files, job submission and interactive tools such as JupyterLab and RStudio.
Faculty can request access as PIs; students usually work through a faculty allocation.
Rung 5: ACCESS, the national supercomputers
At the top of the ladder is ACCESS, the NSF-supported network of national computing resources: large-scale compute, specialized hardware and infrastructure for advanced research.
ACCESS uses credit-based allocations. Projects apply for them, from small exploratory requests to larger, proposal-based awards.
ARCS can help you choose resources, prepare allocation requests and get started on specific systems.
A natural path up the ladder
A typical progression looks like this:
- Start on a laptop. Develop and test the workflow.
- Move to an Iris VM. Get past memory, runtime or storage limits and refine the environment.
- Move to a larger system. Use Hopper, Laguna or ACCESS, depending on the workload’s scale and needs.
Not every project follows this exact path, but the ladder is a useful way to decide.
How ARCS can help
ARCS supports every stage, including:
- finding the right resources for a workload;
- setting up accounts and access;
- developing and scaling workflows;
- guiding students new to research computing;
- pointing to documentation, training and outside resources.
Get in touch
For questions about Iris or research computing at HMC:
- Email ARCS: arcs-l@g.hmc.edu
- Open a Help Desk ticket: helpdesk@hmc.edu
- Nic’s site: cs.hmc.edu/~ndodds
Later posts will cover specific systems, workflows and updates as Iris becomes available.