Overview
The Hopper Cluster is CMC's NSF MRI-funded (Award #2408259) high-performance computing resource, developed collaboratively by CMC and Harvey Mudd College faculty. Named for Grace Hopper, the cluster provides significant computing power for faculty research, intensive data analysis, and complex computational projects that exceed the capabilities of standard workstations.
Hopper sits on the campus-level tier of the research computing ladder — above individual workstations and below national supercomputing centers (NSF ACCESS, DOE NERSC). It fills the gap for researchers who need more than a desktop but less than a national facility, with fair-share SLURM scheduling across research groups, cross-consortium collaboration through the Claremont Colleges, and seamless integration with campus identity systems via OpenLDAP.
The cluster is managed by ITS in partnership with the Hopper Steering Committee, with technical support available at hpc-help@its.cmc.edu.
What HPC Provides
High Performance Computing refers to aggregating computing power to solve complex problems that a single workstation cannot handle. At CMC, the Hopper Cluster delivers:
- Parallel processing — 2,176 CPU cores across 17 compute nodes for parallel CPU workloads
- Large memory capacity — 2 dedicated large-memory nodes with 3 TB RAM each for in-memory analytics
- GPU acceleration — 60 NVIDIA L40S GPUs across 15 GPU-enabled nodes for AI, machine learning, and scientific visualization
- High-speed interconnects — InfiniBand HDR200 between all nodes for low-latency distributed computing
- SLURM workload manager — fair-share scheduling, job prioritization, and resource accounting
- Large-scale storage — 3 PB BeeGFS parallel filesystem
Cluster Specifications
| Resource | Specification |
|---|---|
| Cluster Name | Hopper (NSF MRI Award #2408259) |
| Funding | $1M NSF Major Research Instrumentation grant |
| Compute Nodes | 17 total — 15 GPU-enabled + 2 large-memory |
| CPU Cores | 2,176 cores @ 2.25 GHz |
| GPUs | 60 NVIDIA L40S (4 per GPU node) |
| Memory per Node | 768 GB (15 nodes) or 3 TB (2 large-memory nodes) |
| Storage | 3 PB BeeGFS parallel filesystem |
| Network | InfiniBand HDR200 interconnect |
| Workload Manager | SLURM (Simple Linux Utility for Resource Management) |
| Authentication | OpenLDAP / InCommon Federation (cross-consortium) |
| Software Stack | Python, R, MATLAB, TensorFlow, PyTorch, MPI, CUDA |
| Access | SSH from on-campus or via VPN; email hpc-help@its.cmc.edu |
Where Hopper Fits
Hopper is part of a ladder of research computing infrastructure. No single resource serves all needs — use what is appropriate for the job at hand:
- Personal computers — everyday work, small datasets
- Standalone workstations — moderate computation, single-user
- Hopper Cluster (campus HPC) — parallel processing, GPU acceleration, large-memory jobs
- National supercomputing centers — NSF ACCESS, DOE NERSC for jobs beyond Hopper's scale
Hopper vs. Laguna
CMC faculty also have access to Laguna, the regional HPC cluster at USC. If you're unsure which resource is right for your workload, this comparison can help:
| Resource | Hopper | Laguna |
|---|---|---|
| Location | Claremont McKenna College | Regional (USC) |
| CPU Cores | 2,176 @ 2.25 GHz (128 × 17 nodes) | 2,048 @ 3.1 GHz (128 × 16 nodes) + 512 @ 3.25 GHz (64 × 8 nodes) |
| GPUs | 60 NVIDIA L40S (4 × 15 nodes) | 16 NVIDIA L40S (2 × 8 nodes) |
| RAM per Node | 768 GB (15 nodes) / 3 TB (2 nodes) | 384 GB (16 nodes) / 768 GB (8 nodes) |
| Storage Quota | 50 TB per group | 5 TB per project |
| Scheduling | SLURM (under development) | SLURM + fixed system units |
Hopper's larger GPU count and higher per-node memory make it ideal for AI/ML training and large-memory jobs. Laguna offers higher CPU clock speeds and a more mature SLURM environment.
Governance
The Hopper Cluster is governed by a faculty-led steering committee with representation from CMC and Harvey Mudd College:
- Paul Nerenberg (CMC)
- Shibu Yooseph (CMC)
- Jamie Haddock (HMC)
- Angela Vossmeyer (CMC)
- Bilin Zhuang (HMC)
The committee is currently developing user policies covering acceptable use (e.g., no cryptocurrency mining) and data retention (e.g., not forever). All users must comply with these policies once published.
Getting Access
Verify Eligibility
Hopper access is available to CMC and HMC faculty members and students in their research groups. If you're a faculty member, you're eligible. If you're a student, then please speak with the faculty member you're doing research with.
Request an Account
Use the form below to request an account. ITS will set up your LDAP credentials and grant access to the cluster. If you’d like HPC orientation for you or your team, please make a note of it.
Submit Your First Job
Connect to the VPN using your CMC credentials (OpenLDAP), SSH to Hopper, write a SLURM job script specifying resources, submit, monitor, and retrieve your results.
Best Practices
- Profile your code before submitting large jobs to accurately estimate resource needs and avoid wasting allocation.
- Request only what you need — over-requesting CPU or memory delays your job and reduces throughput for everyone.
- Test on small datasets before scaling to full data — this catches errors cheaply and quickly.
- Keep data organized in clearly labeled project directories; use home directories for scripts and project storage for large files.
- Document your workflows so collaborators and future researchers can reproduce your results.
- Choose the right tier — for smaller workloads, the Quantitative Computing Lab offers high-performance workstations without SLURM overhead. Reserve Hopper for jobs that need its full scale.
Ready to Run Your First Job?
Our Solutions Consultants will walk you through the access request.