GPU Goblin
EN
Beta in preparation

Your idle graphics card can power tomorrow's science

GPU Goblin brings contributors' computers together into a single computing network dedicated to research. With your help, researchers can run their computations and push scientific knowledge forward.

Lend my GPUFor researchers

Linux today, Windows very soon →

How it works

  1. 1

    Install the application

    Sign in with your e-mail address and choose what to lend: processor or graphics card.

  2. 2

    You choose when

    Only while you are away, at night, below a temperature: your machine lends when it suits you, and you can stop it at any time.

  3. 3

    Researchers compute

    Scientists use the lent graphics cards for computations tied to their research.

You have a computer

A gaming PC, a workstation, a lab server: its graphics card sits idle most of the day. Lend that time to research.

  • Only the platform's own code runs, in a sandbox
  • You decide when: away, hours, temperature, memory
  • Stop at once whenever you need the machine
Lend your GPU

You are a researcher

Use GPU Goblin's computing network for your scientific computations from Python: just import the gpugoblin library.

  • Large computations cut into pieces automatically
  • Your data stays on your lab's node by default
  • Results verified before you receive them
For researchers

Built so that you can trust it

Your machine stays protected, and researchers can rely on every result.

No foreign code

Your machine runs GPU Goblin's own published computation code, never a program sent by a researcher.

Sandboxed

Each task runs isolated from your files and network, with limits on memory, disk and processor.

Verified results

A result counts only after an independent recomputation or a match with another contributor's.

You stay in control

Pause, hours, only when you are away, temperature and memory limits: the machine is yours first.

Learn more →

Where the project stands

GPU Goblin is in beta preparation: the grid is being tested before contributors and researchers join in larger numbers. What already works:

  • large matrix computations, and computations described as graphs of tensor operations (written with PyTorch, for example), including ones larger than any single computer, cut into pieces;
  • training machine-learning models, including models larger than any single computer, cut into stages and tiles;
  • computing on processors, on graphics cards through Vulkan (NVIDIA, AMD, Intel) and natively on NVIDIA cards with CUDA;
  • checking every result, either by an independent recomputation or by comparing it with the result of another contributor;
  • the desktop application for Linux, with a console mode for servers and support for machines with several graphics cards.

A Windows version is coming very soon, and an Android version is planned.

Questions people ask first

Is GPU Goblin free?

Yes, for research and any noncommercial use. Lending your computer costs nothing but electricity, and using the grid for research is free. The software is source-available under the PolyForm Noncommercial license.

Will I be paid for lending my computer?

No. Contributors earn credits that count their contribution, but credits have no monetary value and none is promised. GPU Goblin is volunteer computing for science.

Can a researcher run anything they want on my machine?

No. Your machine only runs GPU Goblin's own computation code (matrix products, operations on tensors, training steps of models), inside a sandbox. Researchers describe models and data; they never send programs.

FAQ →

Give science the power it lacks

Sign up in a minute with your e-mail address, install the application, and choose when your machine helps research.

Lend my GPURequest access