GPU rental · submit by task

Plug idle GPUs in
Rent out compute on demand

Video rendering, image generation, audio processing, LLM inference — submit a task and the platform dispatches it to an idle GPU machine, then hands back the output and the logs. Machines at home or behind NAT can join just as easily; no public IP required.

Browse the marketplace Sign in / Sign up
All systems operational Scheduling normal
—
Registered users
—
GPU machines connected
—
Online now
—
Tasks completed
Workload types

Each workload type runs on its own machines

Machines declare their capability tags when they connect, and the platform only dispatches a task to machines whose tags match — a video job never lands on a machine that only serves LLMs.

Loading…
Who it is for

Four kinds of people rent GPUs to get work done

No buying cards, no building machines, no driver or cooling headaches. Submit a task, pay per task, collect the result.

AI short dramas
An episode runs dozens to hundreds of shots. Image generation, voice-over and transcoding all eat compute, and one local card cannot keep up.
  • Batch storyboard image generation, character redraws
  • Voice-over synthesis, lip sync
  • Final-cut transcoding, vertical crops
Designers
Generating the image is only step one. Resizing, upscaling, matting and swapping variants in bulk are what really eat the hours.
  • Batch concept art and key visuals
  • Upscale, matte, cut out, batch template fills
  • 3D rendering, multi-size export
Developers
You want a GPU you can just call, not another dev box to set up for a single run.
  • Model fine-tuning, inference load testing
  • Batch eval-set runs and data cleaning
  • Multi-environment concurrency checks before launch
Researchers
A study runs dozens of parameter sets. A machine sitting idle is waste, and so is waiting in a queue. Compute should be taken on demand.
  • Parameter sweeps and ablation studies
  • Paper reproduction, baseline comparison
  • Long runs that resume from checkpoints
Available worldwide, no region limits
Sign up, submit tasks, read the logs and collect the results — all in the browser, nothing to install. The GPU side is just as unrestricted: home broadband, campus networks and corporate intranets all work. The machine dials out to the platform, so no public IP is needed.
How it works

From topping up to collecting results

1
Add funds
Use a recharge code, or contact an admin to credit your account. Placing an order holds the amount; it is only settled once the task finishes.
2
Submit a task
Pick a machine in the marketplace, write the task title and parameters, submit. This is a demo engine for now — real workload operators open once the client ships.
3
Watch progress
Which machine claimed it, what percentage it is at, what it logged — the console refreshes live.
4
Collect the result
When the task finishes, download the output file. If it fails or times out in the queue, the held amount is returned in full.
Money

Where your money goes, plainly

Held when you order
The full amount is held the moment you submit. It leaves your available balance but has not reached the platform yet — so a run can never fail at the end for lack of funds.
Full refund on failure
Failure, queue timeout and cancelling yourself all return the full amount to your available balance, to the cent.
Reconciled to the cent
Every balance change is written to the ledger. At any moment the balance and the ledger must agree to the cent, and the system checks this itself.