Reaver GPUReaver GPU
API Launch App

Network

Hosting a Node

Turn idle GPUs into recurring income. A node advertises its hardware to the network, runs renter workloads in isolation, and earns per-second rewards settled on Solana.

Hardware requirements

Any CUDA-capable NVIDIA GPU can join, but reward rates and job eligibility scale with the GPU class. The network buckets hardware into standardized tiers so renters can request a class rather than a specific card.

GPUVRAMClass tagTypical jobs
RTX 409024 GBrtx4090Fine-tuning, rendering, batch inference
A10040 / 80 GBa100Training, large-batch workloads
H10080 GBh100Frontier training, high-throughput jobs

Beyond the GPU, a node needs:

  • CPU: 8+ physical cores recommended.
  • RAM: at least 2× total VRAM (e.g. 48 GB for a 24 GB card).
  • Disk: 256 GB+ NVMe for image cache and scratch space.
  • Network: 100 Mbps symmetric or better, stable uptime.
  • Driver: recent NVIDIA driver plus the container toolkit.

Install the node agent

The node agent registers your machine, reports health, and runs sandboxed workloads. Install it and initialize with your operator wallet:

bash
# install the node agent
curl -fsSL https://get.reavergpu.io/node.sh | sh

# register this machine to your wallet
reaver-node init --wallet ~/.config/solana/id.json

Benchmark and go live

Before accepting paid work, a new node runs a verifiable benchmark. The result fixes your advertised class and seeds your reputation score. Re-benchmarking is required after any hardware change.

bash
reaver-node benchmark        # measures compute + bandwidth
reaver-node start --price-floor 0.00012   # lamports/GPU-second
Price floor sets the minimum you will accept. Leave it unset to take the market rate, or raise it during peak demand. The scheduler never assigns work below your floor.

How you earn

You are paid for verified GPU-seconds of completed work. Each job streams signed usage receipts; at settlement, escrow releases to your wallet minus a small protocol fee. There is no payout schedule to wait on — funds land per job.

  1. The scheduler assigns a job that matches your class, region, and floor.
  2. Your agent runs the container in an isolated sandbox and meters usage.
  3. Signed receipts accumulate as the job runs.
  4. On completion, the settlement contract pays you in a single transaction.

Reputation and uptime

Reliable nodes win more work. Reputation rises with successful completions and steady uptime, and falls with failed jobs, timeouts, or dropped connections. Higher reputation unlocks larger jobs and priority in the scheduler's matching order.

Monitoring your node

bash
reaver-node status        # health, current job, reputation
reaver-node earnings --since 7d
reaver-node logs --follow

For production operators, the agent also exposes a Prometheus metrics endpoint on :9095 for utilization, job counts, and earnings.

Next

Hosting income compounds when you also stake behind your node. To understand the sandbox and settlement guarantees, read the Architecture.