Accepting allocation requests
StudentHPC operates Sagittarius, a production HPC cluster available at no cost to enrolled students conducting scientific and mathematical research. No grant line item. No cost share. No industrial queue competing for your nodes.
Reviewed by humans, not an automated scoring system. Denied requests receive written feedback.
About the facility
StudentHPC is a 501(c)(3) nonprofit registered in Colorado. We operate a production high-performance computing cluster and give the cycles away to enrolled students — undergraduate and graduate — doing scientific and mathematical research.
Campus clusters allocate by grant, by department, and by seniority. A first-year working on a genuinely interesting simulation is last in that line, and commercial cloud bills arrive in dollars a student does not have. We exist to remove that constraint entirely.
We take no industrial workloads, sell no compute, and run no paid tier. The facility is funded by donation and grant so that the answer to a student researcher is always yes, or a specific and fixable no.
Research use only
Allocations support scientific and mathematical research. No cryptocurrency, no commercial workloads, no coursework-only usage.
Reviewed by a person
Every request is read by a human reviewer who assesses scientific merit and technical fit. Nothing is scored automatically.
Your data stays yours
We claim no rights over your results, code, or publications. Cite the facility if it helped; that is the entire ask.
System snapshot
Enterprise Dell hardware, a non-blocking clos fabric, and a parallel filesystem — specified below in full so you can size a job before you ever apply.
Compute nodes
Dell PowerEdge R640
Dual-socket 1U nodes running Skylake-SP Xeon Platinum. Every node is identically configured, so a job that runs on one runs on all of them.
Accelerated nodes
Dell PowerEdge C4130
Dense 1U GPU chassis for molecular dynamics, training workloads, and anything that maps cleanly onto CUDA. Requested via a separate Slurm partition.
--nvInterconnect
InfiniBand & Ethernet clos
A leaf-spine clos topology carries MPI traffic on InfiniBand while management and storage ride a parallel Ethernet fabric. Tightly-coupled jobs scale across nodes without contending with filesystem traffic.
Storage
BeeGFS on ZFS
A BeeGFS parallel filesystem striped across ZFS pools. Scratch is fast and purged on a schedule; home is smaller, snapshotted, and meant for code, not data.
| Partition | Intended for | Max walltime | Max nodes/job | Node type |
|---|---|---|---|---|
debug |
Compiling, short interactive tests | 01:00:00 | 2 | R640 |
compute |
Default batch partition | 48:00:00 | 16 | R640 |
long |
Extended single-node runs | 168:00:00 | 1 | R640 |
gpu |
CUDA and accelerated workloads | 24:00:00 | 2 | C4130 |
What runs here
The software stack is preconfigured for the workloads students actually bring us. If yours is not listed, say so in your application — the list grows.
Protein folding, ligand binding, and free-energy work on CPU or GPU partitions.
Aerodynamics, combustion, and multiphase flow across the InfiniBand fabric.
N-body simulation, lattice methods, Monte Carlo, and large-scale data reduction.
Numerical linear algebra, PDE solvers, optimization, and statistical computing.
Genome assembly, alignment, and phylogenetics on high-memory nodes.
Apptainer containers run unprivileged, so an unusual toolchain is a build, not a support ticket.
The process
Three steps, no hidden requirements, and no automated gatekeeping.
Describe the workload, the resources it needs, and the science it serves. Attach a
proposal PDF if you have one. Your .edu address establishes institutional
affiliation and your advisor is named on the form.
A reviewer reads the proposal for scientific merit and checks that the technical request is sized sensibly. If the ask is too large or too small for the science, we say so and tell you what would work.
Approved researchers receive credentials, a scratch directory, and a walkthrough of the scheduler. Declined requests receive written reasoning and may be resubmitted.
Note
Submitting a request does not provision access. Nothing on this site grants a login to any cluster; credentials are issued by a reviewer after approval.
Who can apply
StudentHPC serves enrolled students conducting research under faculty supervision. Undergraduates are as welcome as PhD candidates — the review weighs the science, not your year.
.edu institutional email address
The form takes about ten minutes. You will need your advisor's name and email, and a paragraph describing the workload. Everything else can be refined after review.