Compute for people at RPI.

We hold capacity back for RPI students, graduate researchers and faculty. If your coursework, thesis or grant-funded work needs a GPU bigger than the one in your laptop, this is the way in.

Who this is for

Three groups,
different terms.

We work out which one you are from your application. You do not need to pick correctly — just describe the work as it actually is.

Undergraduate tier

Cost: $0
Node: TROY-1 / SPARK-1
VRAM: 96 / 128 GB
Sharing: None
Storage: Persistent
Owed back: A write-up

Undergraduates

Coursework and personal projects

A monthly allocation of GPU-hours, free, for real academic work: a course project, a competition entry, a research assistantship, or something you are building because you want to. We read every application rather than rubber-stamping them, because capacity is finite.

Graduate researchers

Thesis and dissertation work

Cheaper access with longer allocations, for work that needs a node for weeks rather than an afternoon. If a run has to survive a month of wall-clock time, tell us and we will plan around it.

Faculty

Grant-funded research and teaching

Reduced rates for grant-funded projects, and reserved blocks for courses where every student needs a GPU in the same week. We can also spec and build a machine your grant buys outright.

How we choose

What a good application looks like.

No trick to it. We are looking for people who will actually use the hardware and can say what for.

A specific project. “Fine-tune a 7B model on our lab’s annotated corpus ahead of a workshop deadline” is something we can judge; “I want to learn about AI” is not.
Rough scale. Model size, dataset size, hours you think you need. Guesses are fine, and being wrong is fine. Not knowing yet is a reason to apply, not a reason to hold off.
Why you need a real GPU. Plenty of projects run fine on a free tier somewhere. Saying why yours does not helps us prioritise.
Something back at the end. Write up what you built — a couple of paragraphs and a screenshot is plenty. That is the whole obligation.

If we say no, it is almost always capacity rather than merit, and we will tell you when to try again. In the meantime there is discounted paid access.

Apply

Tell us about the project.

Five fields, about four minutes. We read every one and reply either way.

REVIEWED IN BATCHES REPLY EITHER WAY

Your name

RPI email We use this to confirm your affiliation.

Where you are at RPI

What are you working on? What you are training or running, roughly how big, and what you want out of it.

Scale and timing Rough GPU-hours and any deadlines. “No idea yet” is a valid answer.

We reply to every application.
No mailing list, and we do not pass your details on.

Application received.

We read every one and reply either way. If it is a busy stretch that can take a few days — it has not been lost.