RUNFRA
Compute capacity

The grid · live.

Pooled consumer GPUs running your jobs in parallel. Routed automatically by model, region, and load.

How routing works →
Workers online
Queue depth
Active tasks
Workers busy

Polling cluster…

Routing

Each job is matched to a worker compatible with the requested model and load profile.

Failover

When a worker drops mid-task, its slot is reclaimed and the task retries on another node — invisible to your code.

Refunds

Platform-side failures (timeouts, retries exhausted, every output rejected) refund credits automatically.