Idle GPUs
Nothing changes in your serving stack. Our signed miner runs at full power whenever a GPU is idle and stops the moment your scheduler needs it.
Keep your own customers, API and billing. Peryn mines PRL on your GPUs while they are idle, and your scheduler takes each one back within seconds.
One pod per GPU at a priority below all of your workloads. Your jobs preempt ours, and the GPU is free in about a second.
One job per idle GPU in a preemptible QOS your admin creates. Slurm preempts it like any other job; we never change your scheduler.
The agent gives a GPU back the moment one of your processes starts on it, or when your tools ask with one command.
Weekly time windows in your time zone, and a minimum idle time before we start on a GPU.
How many GPUs we may hold at once, per cluster or per machine, and a temperature limit.
The agent applies your rules with its last copy of them, so a window ends on time even if it cannot reach us.
Nothing changes in your serving stack. Our signed miner runs at full power whenever a GPU is idle and stops the moment your scheduler needs it.
On your serving GPUs a separate miner also uses the gaps between requests. It pauses while any request is unfinished and never runs inside one. Your customers are served as before.
Mining works on numbers our software generates itself, never on your requests, the answers or your models. What we receive: machine metrics (GPU model, phases, how fast each GPU was handed back) and the mining shares. Connections are outbound only and encrypted.
A share of the PRL mined, set in your agreement and taken when it is credited. No setup or sign-up fee.
To your PRL address, through the same pool payouts every Peryn miner gets.
Per day and per GPU: mined, our fee, what you were credited, and the hours each GPU worked. CSV, or print it as a PDF.
Already a partner? Sign in to install, set your rules and download statements.
Tell us about your fleet and your scheduler. We reply within one business day and set up your cloud account.