DS-2 · Dedicated
- AMD Ryzen 9 7950X (4.5 GHz) (16C / 32T)
- 128 GB DDR5 · 2x 2TB NVMe
- 5 Gbps · Unlimited bandwidth
2× 2 TB NVMe: full node with online pruning cycles, 128 GB DDR5 for caches, 16 modern cores — the standard Ethereum box.
Every dapp interaction through a public RPC hands your address, balance queries and mempool intent to an API company. Your own node answers eth_call privately, follows the chain head in real time, and doubles as the backend for a staking validator — no API key, no quota, no logs. Since the merge you run two clients side by side (execution + consensus); since blobs arrived, bandwidth is steady and heavy. The hardware below is sized for that reality, not for 2021 blog posts.
Realistic sizing for a mainnet node in 2026 — chain size, memory, disk and bandwidth. Based on what actually runs in production, not on minimum specs that stopped being true two years ago.
Every plan ships with unlimited bandwidth, NVMe storage and full root access — the three things a node actually consumes. Deploy with a token, pay in the same coin your node secures.
From order to synced node. Commands are copy-paste ready for Ubuntu 24.04 — adapt paths if you run Debian or Rocky.
DS-2 minimum for mainnet. Pick Netherlands or Romania for European peering, Switzerland for the validator you never want subpoenaed.
Key-only SSH, firewall allowing 30303 and 9000, and accurate time sync (chrony) — attestations care about clocks.
Geth, Nethermind or Reth — download, verify checksums, install. One binary, no dependencies.
wget https://gethstore.blob.core.windows.net/builds/geth-linux-amd64-1.16.0.tar.gz
tar xzf geth-linux-amd64-1.16.0.tar.gz && install -m 0755 geth-linux-amd64-1.16.0/geth /usr/local/bin/
Generate the JWT secret, start the execution client, then the consensus client with checkpoint sync — you're following the head in minutes while snap sync backfills state.
openssl rand -hex 32 > /root/jwt.hex
geth --mainnet --http --authrpc.jwtsecret=/root/jwt.hex &
lighthouse bn --network mainnet --execution-endpoint http://127.0.0.1:8551 \
--execution-jwt /root/jwt.hex --checkpoint-sync-url https://mainnet.checkpoint.sigp.io &
Once eth.syncing returns false, point wallets and scripts at your own RPC over SSH or WireGuard. Optionally add a validator client with your staking keys.
geth attach --exec 'eth.syncing'
curl -s localhost:5052/eth/v1/node/syncing | jq
Mid-2026 reality: a full node (execution + consensus) occupies ~1.4 TB after a fresh snap sync and grows ~40–50 GB a month, so 4 TB of NVMe buys years of calm; 2 TB works with regular online pruning. RAM: 32 GB is the practical floor, more becomes cache. CPU: 8+ modern cores; sync is disk-and-network bound, steady state is light. Archive nodes are a different animal — ~3 TB on Erigon or Reth and climbing — which is exactly what the Archive tier mapping is for.
Public RPCs rate-limit you, log you, and go down during the exact volatility you needed them for. A DS-2 running Geth serves your bots, dapps and analytics with zero quota — eth_getLogs over months of history, batched eth_call, WebSocket subscriptions — while staying private. Heavy indexing? DS-3's four NVMe drives and EPYC cores run an Erigon + TrueBlocks stack that would cost hundreds monthly at a commercial API.
A validator is a consensus-client sidecar plus your keys: no extra hardware beyond the full node it rides on. What changes is the stakes — offline means leaking ETH. Bare metal with SLA-backed uptime, ECC memory and no hypervisor between you and the disk is the professional setup. Keys are generated by you, held by you; we rent silicon, not custody. Run one validator or a hundred — the node is the same box.
Blob sidecars made Ethereum nodes chatty: multiple terabytes a month is normal for a well-peered node. That's a rounding error on our unmetered plans and a bill-shock generator anywhere else. And because an RPC endpoint reveals every address you care about, hosting it under a token in Iceland or Switzerland — instead of under an account with your card on file — removes the easiest deanonymization vector in the stack.
Related reading: VPS vs Dedicated Server for Privacy-Critical Workloads · The First Hour of VPS Hardening: A Checklist · Crypto nodes