Skip to main content
Version: dev

Hardware spec

Draft page

For role-specific install commands and deployment patterns, see the Solo sequencer playbook, Full node setup, and Prover setup.

A single-screen view of what each role needs.

Sequencer and full node (single machine)​

ResourceSpec
CPU8 core / 16 vCPU (released 2015 or later)
RAM16 GB
Storage2 TB NVMe SSD with 50,000+ random IOPS
NetworkSymmetric 100 Mbps

The full node and the sequencer use the same hardware. The sequencer differs from the full node in what it does with the machine, not what it needs.

Notes:

  • IOPS matters. SATA SSDs typically come up short. NVMe is required, but cheap NVMe drives without consistent random-write performance can still struggle under sustained load.
  • Disk grows. The activity rate at Alpha plus 18-day blob retention pushes operator-reported disk use into the 1.5 to 2 TB range over time. Provision 2 TB to give yourself headroom.
  • Network bandwidth. Sustained P2P load during epoch boundaries spikes well above the average. Symmetric 100 Mbps is the realistic floor for production.

Staking provider​

Same per-node spec as the solo sequencer. The difference is per-attester multiplication:

  • Each delegated stake position you operate is a separate attester identity.
  • For high availability, run one publisher plus one or more redundant nodes with a coordinated shared keystore.
  • A provider running 10 attester positions across 2 HA-paired nodes is operating roughly 2x the spec above.

Plan for N x per-node-spec plus HA overhead.

Prover​

The prover has a different architecture: one prover node plus one broker plus N prover agents (the workers that do the heavy proof generation).

ComponentSpec per instance
Prover node16 core / 32 vCPU, 16 GB RAM, 1 TB NVMe, 25 Mbps
Prover broker8 core / 16 vCPU, 16 GB RAM, 10 GB SSD
Prover agent (each)32 core / 64 vCPU, 128 GB RAM, 10 GB SSD

The agents do the heavy lifting. Their hardware scales roughly linearly with how many you run:

AgentsCoresRAM
132128 GB
264256 GB
396384 GB
4128512 GB

How many agents do you need​

Guidance: roughly 200 agents per prover at 1 TPS of keccak-heavy transactions (the hardest workload type). Real network load is usually well below that.

If you are provisioning, start with one broker and 2 to 4 agents, watch the broker's queue depth, and scale agents up if the queue grows.

How hardware changes over time​

Three things to plan for:

  1. Disk grows. Blob retention plus accumulated chain state push the disk requirement up.
  2. L1 client co-hosting compounds growth. If you run Geth + Lighthouse on the same machine as the Aztec node (the typical self-hosted RPC pattern), Ethereum's state and beacon-chain data add several hundred GB on top of Aztec's footprint.
  3. V5 will likely shift things. The next major rollup upgrade (estimated mid-2026) includes proof-system optimizations and a new address-derivation scheme. Expect a one-time disk and RAM bump around the cutover.

See also​