Skip to content

Your GPUs earn USD and $PEARL at once.

Serve AI for USD and mine $PEARL on the same GPUs. One command on Slurm, bare metal or any mining OS.

GPUs supported
16
Hopper and Blackwell models.
Hand-back
<1second
to give a GPU back to your scheduler.
Your customers’ data
0
customer requests, answers or models leave the machine.

What one GPU adds each month.

The same GPU earns twice: USD for the tokens, $PEARL for the proofs.

Estimates

Up to, per GPU per month, on top of what your GPUs already earn.

Estimates per GPU per month, up to, per card
GPU Per GPU per month
Hopper
H100 Up to $4,300
H200 Up to $5,200
Blackwell datacenter
B200 Up to $8,500
B300 Up to $8,300
RTX PRO
RTX PRO 6000 Workstation Up to $2,400
RTX PRO 6000 Server Up to $2,400
RTX 50
RTX 5090 Up to $1,600

Estimates at full use: serving Qwen3 30B A3B at our list prices (Llama 3.1 8B on 32 GB cards) with mining while the GPUs are idle, or mining alone where higher. Mining at post-fork launch difficulty, with about 82% of today’s hashrate not carried over (network hashrate by GPU class, prlstats.com ); $PEARL at $1.76 (17 Oct). Results vary.

Your fleet

Serving, at our list prices, USD/month
$9,147
Mining, USD/month
$18,187
Total
$27,333

Estimates before your hardware and power costs; actual results follow demand, the $PEARL price and network difficulty.

What your machines need.

GPUs
Hopper: H100 and H200. Blackwell datacenter: B200 and B300. RTX PRO: RTX PRO 6000 Workstation, RTX PRO 6000 Server, RTX PRO 6000 Max-Q Workstation, RTX PRO 5000, RTX PRO 4500 and RTX PRO 4000. RTX 50: RTX 5090, RTX 5080, RTX 5070 Ti, RTX 5070, RTX 5060 Ti and RTX 5060.
Host
Linux (x86_64 or arm64), Windows, any mining OS, Kubernetes or Docker. NVIDIA driver r580+ (CUDA 13).
Network
Outbound only: HTTPS and our pool’s TLS port. No inbound ports, no public IP.
Where
Your own data centre or any GPU cloud, mainland China included.

One command to set up.

Then watch it connect in the portal.

  1. Kubernetes

    One pod per GPU at a priority below all of your workloads.

  2. Slurm

    One preemptible single-GPU job per idle GPU. The installer prints the QOS settings for your cluster admin and never changes your scheduler.

    On the submit host

    curl -fsSL https://peryn.ai/install | sudo PC_CHANNEL=slurm-fill sh

    On every GPU compute node

    curl -fsSL https://peryn.ai/install | sudo --preserve-env=PC_FLEET_KEY PC_CHANNEL=slurm-idle sh

    PC_FLEET_KEY set by your configuration management (for example exported from a file only root can read), never typed into the line.

  3. Bare metal

    Any Linux with the NVIDIA driver. The machine prints a code to add it to your account.

    curl -fsSL https://peryn.ai/install | sudo PC_RECLAIMABLE=true sh
  4. Docker

    Our agent in one container on each GPU host, next to your own.

  5. Mining OSes

    HiveOS: our agent on every rig.

Your scheduler always wins.

Peryn hands the GPU back in under a second once your scheduler asks. Mining stays inside a strict, measured limit on your response times.

When
Weekly windows in your time zone, after a minimum idle time.
How much
How many GPUs we may use at once, per cluster or per machine.
Kept on your side
The agent keeps your rules, so a window ends on time even when it can’t reach us.
What we receive
Machine metrics and signed claims, over outbound encrypted connections.

Your customers’ data never leaves the machine.

Your data never leaves the machine. Only a signed claim does, and for found blocks, a zero-knowledge proof that reveals nothing about the request.

Nothing leaves
Not in shares, not to our pool, not on the Pearl chain, not in logs.
Leak-tested releases
Every release is leak-tested: we send known text through a GPU and prove none of it leaves the machine.
Never mined
Requests on our Private and Confidential tiers are never mined.

Questions providers ask.

How and when am I paid?

Mined $PEARL goes to your own $PEARL address. A block reward is credited after 100 confirmations. USD for Peryn's API traffic served on your GPUs is paid weekly in a USD stablecoin, under the partner terms. Paid every 2 hours to your $PEARL address once you are owed 1 $PEARL or more.

What do you charge?

Mining: the Duplex fee on the $PEARL your machines mine, in place of the pool fee and with no other fees, set in the cloud agreement. Serving: your share of Peryn’s API traffic is set in the partner terms. Revenue from your own customers is yours, and we never see it.

Partner terms

How fast do I get a GPU back?

Peryn hands the GPU back in under a second once your scheduler asks. Our work runs at a priority below all of yours. On bare metal the agent gives a GPU back as soon as one of your own processes starts on it.

Does any customer data reach Peryn?

Your data never leaves the machine. Only a signed claim does, and for found blocks, a zero-knowledge proof that reveals nothing about the request. What we receive about your machines is their metrics: GPU model, status and hand-back times.

What do you install on my machines?

A signed agent: a systemd service on bare metal, one preemptible job per GPU on Slurm, or one pod per GPU on Kubernetes. It runs our signed runtime, keeps your rules and reports machine metrics over outbound connections. It never changes your scheduler.

Bring your fleet.

Talk to us about your fleet.

We reply within one business day.

We use your name, work email, company, fleet size, GPU types, region and message only to reply and to discuss working with you. We delete the stored enquiry 12 months after your last message, and keep our email thread with you for 2 years after our last reply. Privacy