Idle GPUs,
rented and running in seconds.
JupyterLab & PyTorch on a consumer-GPU pool, instantly. Not a datacenter — a public GPU network running on personal PCs.
Sign up for a $0.40 trial credit · passwordless email login · from $0.30/hr
These GPUs are running right now
The pool is small and curated for now. Not a hyperscaler's marketing number — only what's actually powered on this very moment.
※ The pool is small for now — a few personal PCs (RTX 4060 Ti·4070 SUPER). Rent directly from the dashboard after you sign in. The numbers above are a live control-plane snapshot, so they change on every refresh.
See it actually run right now — try the free background-removal demo (no signup) →
Open models you can run right now
Different GPUs run different models, so we match each one to the GPU's power. A bigger GPU runs a bigger model, a smaller one runs a lighter one. Everything here we've run ourselves and checked for speed and function-calling.
Honest — the good and the limits.
Start instantly; if you're stuck, a human answers. And we don't hide what we can't do.
Instant after sign-up
Launch JupyterLab with no card. A trial credit = 60 min free; if you like it, from $0.30/hr (per GPU/VRAM), per-second billing.
Yields when the owner returns
When the provider hits a key, the workload pauses instantly (~1ms, process-internal). But if a node drops, the session restarts on another node — in-progress memory isn't preserved (not live migration).
Isolation, fully disclosed
Isolation is process-level — on a personal PC it isn't fully separated from the host (confidential computing needs H100-class hardware; unsupported).
How it works
Idle most of the time, runs when needed, yields when the owner returns. A simple promise, executed precisely.
Share idle GPUs
Provider PCs lend their GPU to the network only while idle. Usage is settled and 70% goes back to the provider.
Process-isolated execution
Workloads run as a separate process. But on a personal PC it isn't fully isolated from the host, so don't upload sensitive data. See the isolation matrix →
Yields on owner return
When the provider uses keyboard/mouse, the workload pauses. Yield latency measured at ~1ms (process-internal, excluding network/scheduling).
Start now
Sign up for a $0.40 trial credit — 60 min of free RTX GPU. To use more, from $0.30/hr.