SpareCycles splits a Blender render, a training run or a data job into chunks and runs them on lab machines and student laptops that would otherwise sit idle. Contributors earn credits for every chunk their machine finishes.
You upload files. The control plane decides what they are, packages them into a container, cuts the work into chunks and hands each chunk to a machine on the network.
Drop a .blend, a .py or .ipynb, or an archive of your project.
The pipeline works out the workload type, framework and resources. Gemini fills in what heuristics miss.
A pre-verified image is matched or built, after a scan for malicious code and miners.
Frame ranges for renders, ranks for training, shards for data, domains for simulations.
Each chunk goes to the best-ranked online node. LAN peers are grouped for multi-node jobs.
Outputs are merged into one video, model, dataset or result set you download.
Machines leave. If a node stops sending heartbeats, the watchdog lowers its reliability score and requeues its chunks on another machine.
| Workload | Input | Split by | You get back |
|---|---|---|---|
3D render Blender | .blend | frame ranges | frames + assembled video |
ML training PyTorch | .py / .ipynb | ranks (DDP or Local-SGD) | weights + checkpoints |
Data processing Python | script + data archive | shards | merged dataset |
Simulation OpenFOAM / MPI | case archive | domains | merged results |
Use our sample scene: a 72-frame animated scene that renders in about two minutes on two laptops.
sparecycles-sample.blend · 1.4 MB · 640×360
Sign in with Google, upload your files and watch every chunk move across the network live. When the last one lands, download the assembled output.
Submit a jobInstall the SpareCycles desktop app and switch it on. It runs chunks in local Docker containers, uses your GPU if you have one, and earns credits for each chunk it completes.
Contribute a machine