Provider infrastructure

GPU nodes

Provider portal

Supply capacity

Put a GPU on the network

Prism pays 90% of confirmed usage to the payout wallet you nominate, settled in USDG against an onchain receipt. Joining takes a bond in PRISM, held by the registry while the node serves and returned when it retires.

The bond scales with the rate a node charges. Ask the registry what yours costs before committing anything: the dry run below signs the registration and reports what would happen without spending.

Host requirements

Infrastructure baseline

  • Ubuntu 24.04 x86-64
  • NVIDIA GPU and driver
  • IOMMU/VFIO enabled
  • containerd and Kata runtime
  • Outbound mTLS tunnel

A host reached over the outbound tunnel and running the isolated stack is published above open capacity, which is the tier renters can require for workloads they will not place on a machine that can read guest memory.

Provider onboarding

Node registration

prismd preflightprismd create-identity --path /var/lib/prismd/device.jsonprismd register --identity /var/lib/prismd/device.json --rpc-url https://rpc.mainnet.chain.robinhood.com --registry 0xa7Ca8e43c599b978095c391bd018A35BA6e7B71D --rate-per-second 222 --dry-runprismd enroll --identity /var/lib/prismd/device.json …prismd certificate --identity /var/lib/prismd/device.json …

Enrollment is permissionless: the control plane accepts any node whose device key signs the request and whose bond the registry confirms. The installation guide covers the host baseline, the bond and the services in full.

Already running a node?

Sign in with the operator or payout wallet used during enrollment to see its inventory, certificates and settled earnings.