Hardware requirements
A Prova prover doesn't need exotic hardware. The protocol is disk-bound, not CPU-bound — no PoRep, no SNARKs, no GPU.
Minimum
- CPU: 4 cores (2.0 GHz+)
- RAM: 8 GB
- Disk: anything ≥ 1 TB. SATA SSD or NVMe preferred. Spinning rust works for cold storage.
- Network: symmetric 100 Mbps. Real upload, not consumer-asymmetric.
- Connection: real IP, real DNS. No CGNAT.
- OS: Linux (Ubuntu 22.04+, Debian 12+, RHEL 9+ tested). macOS works for dev. Windows untested.
Recommended (10 TB tier)
- CPU: 8 cores
- RAM: 32 GB
- Disk: 10 TB NVMe or 4× 4 TB SSD ZFS RAID-Z1. ECC where possible.
- Network: 1 Gbps symmetric. Multi-homed if you can.
- Power: UPS. Single power loss during a proof = missed proof = slash.
At scale (100 TB+)
- Server-grade hardware. Hetzner SX-line, OVH bare metal, used Dell R7xx with shelves.
- ZFS or btrfs with checksumming and scrubbing.
- Two upstream providers (BGP if you're feeling spicy).
- Cooling. Real cooling.
What disk shape
| Disk | Good for | Notes |
|---|---|---|
| NVMe | Hot tier, fast retrievals | Most expensive per byte |
| SATA SSD | Default. Best balance. | Sweet spot for now |
| SAS HDD (7.2k or 10k) | Cold tier, cheap bulk | Slower retrievals → may lose deals to faster provers |
| SMR HDD | Don't | SMR's write amplification kills proof latency |
You can mix tiers. The prover daemon supports a "hot" tier (recent + popular pieces) and a "cold" tier (rarely-retrieved). Caching strategies live in prover/pkg/store/.
What CPU
PDP proofs are SHA-256 on a few KB at a time. Any modern CPU does this in microseconds. The CPU bottleneck (if any) is TLS termination + HTTPS retrieval throughput — i.e., serving clients, not generating proofs. Modern AES-NI CPUs handle gigabit-line-rate HTTPS without breaking 5% utilization.
What network
The two metrics that matter:
- Upload bandwidth. You're serving every retrieval. A 1 Gbps link means ~125 MB/s peak retrieval throughput. If your provers are popular, more = more revenue.
- Latency. Clients route to the fastest healthy prover. A prover in São Paulo serving Norwegian users will lose to one in Frankfurt. Place your hardware where the demand is.
Geographic placement
If you're starting with one node, put it where:
- Land and electricity are cheap (Nordic countries, Iceland)
- Latency to ENS / NFT / web3 traffic is good (EU + NA East coast)
- Regulatory environment is friendly (Norway, Switzerland, Singapore)
If you're scaling to 5+ nodes, spread across continents. Prova's marketplace prefers geographically-diverse redundancy, which means cross-continent provers see preferential placement.
Stake (the hidden cost)
Your biggest hardware-equivalent expense is the USDC stake you must lock to commit capacity. The current ratio is ~$50 stake per TiB committed (configurable per the ProverStaking.sol parameters; check before registering).
For a 10 TB prover:
- Disk: $1,000-2,500 (one-time)
- Network: $50-100/month
- Stake: ~$500 USDC locked (refundable when you exit cleanly)
- Slash exposure: up to your full stake if you fail badly
Plan accordingly.