Real GPUs for your class
Give every student their own GPU for AI and data-science coursework. You share one link, hand out credits and set the limits; they open JupyterLab in the browser. Nothing to install, no lab to maintain, and you pay only for the minutes the machines run.
How a class works
- Create the class. Sign up, open Class in the console and name your class. You get a join link.
- Share the link. Students sign up with Google or an email address and join in one click. Each student has their own account and their own machines.
- Hand out credits. Give each student credits from your account, and take back what they do not use. Students can also add credits of their own.
- Set the limits. A daily and a monthly spending limit for each student, the software they may use, and the most a machine may cost an hour.
- Teach. Students rent a GPU and open JupyterLab, ComfyUI or a private chat model in the browser. You see each student’s credits, spending and running machines, and can stop any of them.
What it costs
Worked out from today’s board: the 30 cheapest GPUs with at least 12 GB of memory (such as the GTX TITAN X, RTX A2000, RTX 3060, Titan Xp), enough for most AI coursework. Billed by the second only while a machine runs; a stopped machine pays only for its disk, at the rate shown before renting.
| For | Cost today |
|---|---|
| One student-hour on a 12 GB+ GPU (typical) | USD 0.12 |
| A 2-hour lab for 30 students | USD 7.04 |
| One student’s 10 hours of project work | USD 1.18 |
| A semester: 12 weekly labs of 2 hours, 30 students | USD 84.48 |
Nothing runs away
- Each student’s daily and monthly limit: a warning at 80 %, and their machines stop at 100 %. Students cannot raise the limits you set.
- Only the software you allow, and only machines under the hourly price you choose.
- A timer on each machine, and a warning when a GPU has done no work for an hour.
- Stop any student’s machines from your screen. Their disks are kept.
- Take back unused credits at any time, or remove a student from the class.
What students get
- JupyterLab with PyTorch or TensorFlow on a real NVIDIA GPU, in the browser.
- ComfyUI and Stable Diffusion for image work, Ollama with Open WebUI for chat models, Whisper for speech.
- No SSH key and nothing to install; SSH is there for those who want it.
- Pick for me: students choose the job, and we choose the machine and show the whole price first.
- Their own account and machines: classmates never see each other’s work.
Questions
Do students need a payment card?
No. You give them credits from your account. Students can also add credits of their own if they want more.
Do students need to install anything or make an SSH key?
No. Every machine opens in the browser: JupyterLab, and the app of the software chosen. An SSH key is optional.
Can I see what my students do?
You see each student’s credits, what they spent today and this month, their limits and their running machines, and you can stop those machines. You do not see their files.
What if a student leaves a machine running all night?
Their daily limit stops it, a machine timer can stop it, the idle warning tells them, and you can stop it yourself from the Class screen.
How does the college pay?
By card on Stripe’s secure payment page, in your account currency. If the college wants to pay for the whole class as an institution, write to support@powerpod.si.
Which GPUs can students use?
Anything on the board, from 12 GB cards that are plenty for coursework to RTX 4090, A100 and H100 cards for research. You can cap the hourly price for your class.
Is there a discount for colleges?
Write to support@powerpod.si to have your college verified. Verified classes can receive bonus credits on top-ups.