Invite-only GPU sharing for a student AI club

Shared GPU power.

For the VIBE club.

A new angle on compute.

Unlock the chip.

A new perspective on training.

Run approved AI workloads on the club's own GPU. Free to members, within fair limits. See how it works

What you get

  • Private

    Your data and results stay on hardware the club controls. Nothing goes to an outside AI service.

  • Free within quotas

    No bill and no credits. 2 jobs at once, 10 a day and 500 MB of storage per member.

  • Measured

    Every job records its time and peak memory.

  • Nothing to install

    Pick a ready-made workload in your browser and press Run.

No installs. No bills.

Succeeded Small image classifier 2 rounds · 23 s · Quadro P400

How it works

How a job runs

  1. Pick an approved workload

    Choose from the list the admin trusts. Each one runs from a pinned container image.

  2. Set it up

    Rounds, batch size, seed, or your own CSV table. Every setting is checked before the job is queued.

  3. Watch it train

    The live log refreshes every 5 seconds, with a round counter and a loss chart.

  4. Keep the results

    Download metrics.json and the trained model.pt from the job page.

job · small image classifier
worker desktop-gpu-1 claimed job
device cuda:0 · Quadro P400
round 1 of 2 · done
round 2 of 2 · done
test accuracy 98.38%
Succeeded · 23 s
Live log
Loss per round · illustration
Quadro P400 · peak memory 463 MB
of 2048 MB
GPU use
metrics.json ↓
model.pt ↓
steps.csv ↓
Results

Limits

Why it's safe to share

Sealed containers

Jobs run with no network, read-only files, a non-root user, and memory and time limits.

Fair limits

Per-member limits keep the queue moving for everyone.

Honest numbers

The site never says a GPU job worked unless the job's own output proves it.

2Jobs at once
10Jobs a day
500 MBStorage per member

Let's get training.

GPU1 to 5 rounds

Small image classifier

Train a handwritten-digit recognizer and watch accuracy rise each round.

98.38% after 2 rounds, 23 s on our Quadro P400.

GPU

PyTorch GPU check

Proves the GPU is really doing the work.

1024 × 1024 matrix multiply: 4.2 ms on the GPU, 21.1 ms on one CPU core (5.1×).

CPUYour CSV

Train on your own table

Upload a CSV and train a small model to predict one column.

The pilot

Join the pilot

The first 3 to 5 VIBE members try it in person. Want in? Message the admin on the VIBE Discord.