On this page 10 sections
Solana is unusually demanding for a blockchain node: Anza’s current requirements start at 12 fast cores, 256 GB of RAM and several NVMe drives for a validator, and 512 GB of RAM for an RPC node with full account indexes. This guide turns those requirements into concrete choices of CPU, memory, disk layout, network and tuning, and notes what is changing in 2026. It covers hardware only and is not staking or financial advice.
The official baseline
Anza, which maintains the Agave validator client, publishes the reference requirements and encourages operators to do their own performance testing. As of September 2026 they read as follows.
| Component | Validator | RPC node, in addition |
|---|---|---|
| CPU | 12 cores / 24 threads or more; 2.8 GHz base clock or faster; SHA extensions; AMD Gen 3 or newer, or Intel Ice Lake or newer; AVX2 for the official binaries; AVX-512F helpful | 16 cores / 32 threads or more |
| RAM | 256 GB or more; ECC suggested; a motherboard that takes 512 GB suggested | 512 GB or more for all account indexes |
| Disks | A PCIe Gen3 x4 NVMe drive or better for each of: accounts (1 TB+, high TBW), ledger (1 TB+) and snapshots (500 GB+); OS optional (500 GB+, SATA acceptable) | A larger ledger disk for longer history; accounts and ledger should not share a disk |
| Network | Static public IPv4; 1 Gbps symmetric for an unstaked node; at least 2 Gbps symmetric for a staked node, 10 Gbps recommended | Same |
| Software | Ubuntu 24.04, Anza’s build and run platform; Docker not recommended for mainnet | Same |
CPU: clock speed first
Anza is explicit that a higher clock speed is preferable to more cores. Its setup guide links a Proof of History hash rate below the cluster target to CPUs under the 2.8 GHz base clock, and recommends the performance CPU governor so that frequency scaling does not pull clocks down.
Instruction sets are rarely the problem on current hardware. Every AMD Zen 4 and Zen 5 CPU meets Anza’s instruction-set requirements and adds the AVX-512 support it calls helpful. AVX-512 matters even more for Firedancer, whose signature verification uses a custom AVX-512 ED25519 implementation.
Base clock is the real filter. Many high-core-count server CPUs sit below 2.8 GHz: the EPYC 9454P is rated at 2.75 GHz base with a 3.65 GHz all-core boost, and the 192-core EPYC 9965 at 2.25 GHz. The EPYC 9355P, at 3.55 GHz, and the 9555P, at 3.2 GHz, clear the bar with margin.
RAM: validators vs RPC nodes
A validator needs 256 GB as a floor. RPC nodes need more because of account indexes: in-memory indexes that make queries such as getProgramAccounts and getTokenAccountsByOwner fast. Agave supports three, enabled with --account-index: program-id, spl-token-mint and spl-token-owner. With all three enabled, Anza’s guidance is 512 GB or more.
ECC memory is suggested for both roles. It corrects single-bit errors and turns many others into detected, logged events instead of silent corruption, a sensible property for a machine that holds consensus state. Anza also notes that you typically don’t want one machine to be both a consensus node and a full RPC node, because of resource constraints.
NVMe layout: accounts, ledger, snapshots
Disk layout matters as much as disk size. Anza recommends a separate NVMe device for each of the accounts database, the ledger and snapshots. Anza lists high write endurance as a requirement for the accounts drive and only a suggestion for the others, which tells you where the writes land. The setup guide asks for at least 2 TB of space for the ledger.
- Two drives. Accounts alone on one drive; ledger, snapshots and the OS on the other. Anza allows the OS on the ledger disk, though it notes the ledger performs better on its own.
- Four drives. Accounts, ledger, snapshots and OS each on their own device. For an RPC node, keeping accounts and ledger apart is a requirement, not a preference.
- Endurance. TBW, total bytes written, is a drive’s rated write endurance. Check the rating before deployment and track wear afterwards, for example with
nvme smart-log, whose “percentage used” field estimates how much of that endurance is spent.
Network: 1, 2 or 10 Gbps
Anza’s networking guidance is tiered: 1 Gbps symmetric is sufficient for an unstaked node, while a staked node needs at least 2 Gbps symmetric, with 10 Gbps of available bandwidth recommended for stable operation. A static public IPv4 address is required, ports 8000–8030 must be reachable over TCP and UDP, and Anza advises against running behind NAT.
Monthly volume depends on stake, role and client load. Tens of terabytes per month is a realistic order of magnitude for a busy node before any RPC responses to users, which is a real line item on a metered plan and none on an unmetered port. An RPC node also needs headroom for client traffic on top of cluster traffic, which makes 10 Gbps the sensible default for that role.
OS tuning pointers
Anza publishes the kernel settings a validator needs and warns that the node may not start without them:
# /etc/sysctl.d/21-agave-validator.conf
net.core.rmem_max = 134217728
net.core.wmem_max = 134217728
vm.max_map_count = 1000000
fs.nr_open = 1000000- Load them with
sudo sysctl -p /etc/sysctl.d/21-agave-validator.conf. - Raise file and locked-memory limits in the service unit:
LimitNOFILE=1000000andLimitMEMLOCK=2000000000. - Set the CPU frequency governor to
performance. - Mount accounts and ledger on separate NVMe devices, as described above.
- Run on bare metal: Anza does not recommend Docker for mainnet and warns that cloud VMs demand significantly more operational expertise.
Firedancer, Frankendancer and Alpenglow
Solana now has a second validator implementation. Frankendancer, which runs Firedancer’s networking and block-production code alongside Agave’s runtime, went live on mainnet in September 2024. The fully independent Firedancer client went live in December 2025 after about 100 days in production on a handful of validators, according to The Block. In May 2026, Jump’s lead engineer told CoinDesk the team did not want everyone to run it yet, before full security audits; it had recently completed a public audit competition with a $1 million bug bounty pool.
Firedancer’s documentation puts Frankendancer’s hardware needs at least at Agave’s level: a 24-core CPU above 2.8 GHz and 256 GB of RAM as a minimum, and a 32-core CPU above 3 GHz with AVX-512, 512 GB of ECC RAM and separate accounts and ledger disks as the recommendation. It must be built from source, and binaries are optimized for the CPU they are built on.
Consensus is changing as well. Alpenglow replaces TowerBFT voting with votes sent directly between validators; its first phase, Votor, ships in Agave 4.3. When we last checked Solana’s official upgrade page in September 2026, mainnet activation was expected in Q3 2026 and had not yet happened. Vote transactions will leave blocks, so re-check bandwidth and CPU assumptions after activation instead of sizing for today’s vote load alone.
Concentration and geography
Where a node runs matters to the network, not only to its operator. In November 2022, when Hetzner enforced terms prohibiting crypto activity, more than 1,000 Solana validators went offline and more than a fifth of stake became delinquent, The Block reported. The network itself appeared unaffected, but the episode showed the risk of stake clustered at one provider.
The Solana Foundation Delegation Program now encodes that lesson. Since 1 May 2026, participants must run in a data center holding no more than 15% of network stake, on an ASN and hosting provider holding less than 25%; validators.app publishes the current figures. Geography also affects performance: a validator far from most of the stake sees more latency on votes and blocks, and an RPC node belongs close to the applications that query it. Check concentration before choosing a city, not after.
Mapping the requirements to real servers
| Role | Configuration | Fit against Anza’s guidance | Price, delivery, locations |
|---|---|---|---|
| Staked validator | EPYC 9355P: 32 cores at 3.55 GHz base, 384 GB DDR5 ECC, 2 × 3.84 TB NVMe Gen5 | Clears the clock, core and RAM bars with headroom; with two drives, accounts get one and everything else the other | $519/month, 24 h; AMS, FRA, NYC, TYO |
| Budget validator or unstaked node | EPYC 9454P: 48 cores at 2.75 GHz base, 3.65 GHz all-core boost, 256 GB DDR5 ECC, 2 × 3.84 TB NVMe | Meets the 256 GB floor exactly; base clock is 50 MHz under the 2.8 GHz guideline; no room for account indexes | $489/month, about 15 min; AMS, FRA, NYC, LAX, SIN |
| RPC node with all account indexes | EPYC 9555P: 64 cores at 3.2 GHz base, 768 GB DDR5 ECC, 4 × 3.84 TB NVMe Gen5 | Well above the 512 GB index guidance; four drives allow the full layout | $949/month, 24 h; AMS, FRA, NYC, TYO |
Our Ryzen configurations top out at 192 GB of RAM, below the 256 GB baseline, so we do not suggest them for mainnet validators or RPC nodes.
Every PowerDedi server includes a 10 Gbps unmetered port and IPMI/KVM access, all EPYC configurations use ECC memory, and you can pay in SOL, or in Bitcoin, Ethereum, Monero or USDT on TRON. Current stock is on the blockchain nodes page, with full specifications in the AMD EPYC range.
Sources
- Agave validator requirements — Anza
- Setup an Agave validator — Anza
- Setup an Agave RPC node — Anza
- Getting started — Firedancer documentation
- Jump’s “Frankendancer” validator client is live on Solana mainnet — CoinDesk, 20 September 2024
- Jump Crypto’s Firedancer hits Solana mainnet — The Block, 12 December 2025
- Jump Crypto’s Firedancer is taking a slow and steady approach — CoinDesk, 16 May 2026
- Alpenglow — Solana Foundation
- Delegation criteria — Solana Foundation
- Data center concentration — validators.app
- 1,000 Solana validators go offline as Hetzner blocks server access — The Block, 2 November 2022
