OURO Grid · Nodes
Plug your idle GPU into the evolution network.
Nodes run evaluation and sampling tasks and settle in $OURO by tasks completed and verification pass rate. Your device only touches de-identified task shards, never full weights or private eval sets.
Hardware tiers
| Tier | GPU | VRAM | Upload bandwidth | Eligible tasks |
|---|---|---|---|---|
| Light node | No GPU needed | — | ≥ 20 Mbps | Data preprocessing, result verification |
| Standard node | RTX 3060 / 4060 or better | ≥ 8 GB | ≥ 50 Mbps | Candidate evaluation, inference sampling |
| Advanced node | RTX 4070 Ti / 4080 / 4090 / 5090 | ≥ 16 GB | ≥ 100 Mbps | + collective RL sampling, small-model fine-tuning |
| Pro node | A100 / H100 / H200 etc. | ≥ 40 GB | ≥ 1 Gbps, latency < 150 ms | + low-communication training shards, redundant recompute |
| Regional anchor | Data-center grade, multi-GPU interconnect | — | Dedicated line | Arbitration, holdout blind tests, TEE, weight-level tasks (staking and review required) |
Windows, macOS (Apple Silicon can run evaluation tasks) and Linux; multi-GPU and small data-center modes.
How earnings work
Settlement = tasks completed × verification pass rate × task weight. New nodes have a trial period; quota grows with reliability score. The reference is the market rental price of the same GPU (RTX 4090 about $0.3–0.4/hour). No fixed return is promised; no network demand means no emission.
What your device cannot see
Not the full model weights, not the private eval sets, not other nodes' results. Task shards are de-identified before dispatch, so a single node cannot reconstruct the full context. The client is open source and auditable.
How cheating is prevented
Activation-hash spot checks against skipped compute, multi-node redundancy against false reports, holdout blind tests against overfitting. Forged results confirmed by anchor recompute lose quota and stake; reliability scores are public on-chain.
How tasks are assigned
The scheduler assigns by task type, node capability, historical reliability, location and live price: latency-tolerant evaluation follows price, critical tasks follow reliability, and redundant recompute is deliberately spread across regions and operators. Network health (nodes per region, completion rate, evaluation agreement rate, average cost per evolution round) is public on-chain and on a dashboard.
Regional anchors
The first anchors are in Singapore and Johor Bahru, handling arbitration, holdout blind tests and baseline capacity. Later anchors are co-built with communities and partners; candidate regions are Northeast Asia, Europe and the central US. Anchor operators stake $OURO and their rulings are cross-checked by other anchors. Teams with data-center resources can tick "interested in co-building a regional anchor" on the waitlist form.
FAQ
Do I need to buy $OURO first?
On the testnet you stake test tokens, no purchase needed. After mainnet the stake is small and exists to prevent Sybil attacks.
Will it hog my computer?
You can set online hours and a GPU usage cap, and it yields automatically when you are using the machine.
Which regions cannot take part?
Running a node is unrestricted; $OURO-related features are limited in restricted jurisdictions by geography and KYC rules, see the risk statement.
When can I install it?
It ships with the PoE testnet in 2027 Q2; the waitlist is notified in registration order.