Serving first. Mining in every gap.
Our orchestrator routes each AI request to a GPU, fills the idle time with mining, and hands any GPU back the moment its owner asks.
Three decisions, for every GPU.
Mining during serving is switched on only where we measured it adding under 5% to response times (under 5 ms at low load), so your customers are served as before. Available with our datacenter runtime.
- Where a request runs
- On a healthy GPU that serves the model, inside the countries the customer allows.
- When a GPU mines
- Whenever it is idle, and with our datacenter runtime in the gaps while it serves.
- When a GPU goes back
- The moment its owner’s scheduler asks for it, whatever it was doing.
Your GPU, back when you ask.
Peryn hands the GPU back in under a second once your scheduler asks.
It works inside Kubernetes, Slurm or bare metal: how it runs in your scheduler. Mining on your own rigs instead? Mine with Peryn.