--- title: "Hyperliquid Node Requirements: Specs, Tokyo, Ports (2026)" description: "Hyperliquid node requirements: 16 vCPU, 128 GB RAM and 500 GB SSD for a non-validator, Ubuntu 24.04, ports 4001–4002, Tokyo, and 100 GB of logs a day." url: "https://powerdedi.com/blog/hyperliquid-node-requirements" last_updated: "2026-10-15" publisher: "PowerDedi (https://powerdedi.com)" --- # Hyperliquid node requirements: hardware, Tokyo location and disk use What a Hyperliquid non-validating node or validator needs, from the official node repository: vCPUs versus physical cores, 128 GB of RAM, Ubuntu 24.04, gossip ports, why the README says Tokyo, and how to prune about 100 GB of data a day. By PowerDedi Engineering · October 15, 2026 · 9 min read **Share this article** A Hyperliquid non-validating node needs 16 vCPUs, 128 GB of RAM and a 500 GB SSD; a validator, 32 vCPUs, 128 GB and 1 TB. The node runs only on Ubuntu 24.04, needs public gossip ports 4001–4002, writes about 100 GB of logs a day by default, and the official README recommends Tokyo for the lowest latency. This guide is based on the hyperliquid-dex/node repository and Hyperliquid’s documentation as of October 2026. It covers hardware, location and operations, not trading: running a node does not change who may trade on Hyperliquid or any obligation you have where you live. ## The short version | Requirement | Non-validating node | Validator | | --- | --- | --- | | CPU | 16 vCPUs | 32 vCPUs | | Memory | 128 GB (raised from 64 GB in May 2026) | 128 GB | | Storage | 500 GB SSD | 1 TB SSD | | Operating system | Ubuntu 24.04 only | Ubuntu 24.04 only | | Ports | 4001 and 4002, open to the public | 4000–4010 open to other validators (4001–4006 in use today) | | Location | Tokyo, “for lowest latency” | Tokyo, and 200 ms two-way latency to at least a third of the stake | | Data written | About 100 GB of logs a day with default settings, in `~/hl/data` | Same | | Who can run it | Anyone: permissionless | Anyone, but the active set is the top 27 by stake, with 10,000 HYPE of self-delegation | ## Non-validator or validator: which one you need Most operators want a **non-validating node**. It follows the chain from peers and writes the L1 data to disk, and it can serve two local endpoints: an EVM JSON-RPC (`--serve-eth-rpc`, on `localhost:3001/evm`) and an info server (`--serve-info`, on `localhost:3001/info`) that answers a subset of API requests from local state. The README notes that a local info server “can help with rate limits and reduces trust assumptions on external operators”. Optional flags stream trades, fills, order statuses, raw book diffs and other events to files. Hyperliquid’s docs put it plainly: “running a non-validating node is permissionless.” A **validator** is a different commitment. Hyperliquid’s docs say the active set is “the top twenty-seven by stake”, and each validator needs a 10,000 HYPE self-delegation, locked for one year, to become active. A newly registered validator, or one that changes its IP address, starts jailed and must unjail itself once its consensus logs look healthy. The README also asks validators to “dedicate most of their machine’s resources to run consensus”, to serve the public through up to two sentry nodes that they run themselves, and to run the rate publisher for aligned quote assets on a separate machine. In practice that is several servers, not one. One detail dates older guides: on May 9, 2026, the README’s machine specs raised the non-validator’s memory from 64 to 128 GB. Anything that still says 64 GB predates that change. ## What “16 vCPUs” means on a dedicated server The README sizes nodes in vCPUs, a cloud unit. On AWS, for example, “each thread is represented as a virtual CPU (vCPU)”: an m5.xlarge has two cores with two threads each, so four vCPUs. Sixteen vCPUs therefore usually means eight physical cores with simultaneous multithreading. A dedicated server is described in cores and threads. A Ryzen 9 9950X has 16 cores and 32 hardware threads: twice the non-validator figure, and equal to the validator’s 32 vCPUs. An EPYC 9355P has 32 cores and 64 threads. Hyperliquid’s latency guide sets a higher bar for nodes that feed trading systems: “at least 32 logical cores, 128 GB RAM, and 500 MB/s disk throughput”, adding that more cores can reduce latency “because blocks will be faster to execute”. The same page calls its suggestions informational, “not trading advice”, with no guarantee of any latency or outcome. Read the README figure as the floor and the latency guide as the target if response time matters to you. ## Where to run it: why Tokyo The README is direct: “For lowest latency, run the node in Tokyo, Japan.” Three other published facts point the same way: - **The Foundation’s peer.** The Hyper Foundation runs a non-validating node as a reliable, low-latency peer, and its docs say it “runs in apne1-az1 on AWS”. AWS lists apne1-az1 in ap-northeast-1, its Asia Pacific (Tokyo) region; an AZ ID refers to the same physical location in every AWS account. - **The seed list.** Of the 25 community root peers listed in the README (revision of August 10, 2026), 20 are in Japan, two in South Korea, two in Singapore and one in Germany. - **The jailing rule.** To avoid being jailed, the README recommends that validators achieve “200ms two-way latency to at least one‑third of validators by stake”. We do not publish latency figures for our own Tokyo servers: one measurement from one machine to one peer at one moment says little about your setup. Measure from the server you intend to use instead: ```bash # Root peers currently available to non-validators curl -X POST --header "Content-Type: application/json" --data '{ "type": "gossipRootIps" }' https://api.hyperliquid.xyz/info # Round-trip time from your server to one of them (replace ROOT_IP) ping -c 20 ROOT_IP ``` Once the node runs, the README points to `~/hl/data/node_logs/status/` for diagnosing latency or connectivity issues. Tokyo is about latency, not correctness. Nothing in the README says a node elsewhere will fail, and its seed list includes peers in Singapore and Germany. If you archive data or run analytics that are not time-sensitive, location is a secondary choice. Direct access to the Foundation’s peer is also restricted: the docs require 10,000 staked HYPE, maker rebate tier 1 or above, 98% uptime and compliance with legal and sanctions conditions. Everyone else can buy peering from the infrastructure providers the docs list. ## Disk use: about 100 GB a day, and how to prune it The README’s warning is the most practical line in it: “With default settings, the network will generate around 100 GB of logs per day, so it is recommended to archive or delete old files.” That figure has been in the README since May 2025, so treat it as an order of magnitude and measure your own. At that rate an unpruned 500 GB disk fills in about five days. Everything lands under `~/hl/data`: - `replica_cmds/`: blocks parsed as transactions, by start time, date and height. - `periodic_abci_states/`: a state snapshot every 10,000 blocks. - `node_logs/`: consensus and status logs, useful for debugging. - `visor_child_stderr/`: crash logs from the node process. - Optional outputs, one folder per `--write-*` flag: trades, fills, order statuses, raw book diffs, miscellaneous events and others. The README warns that order statuses and raw book diffs “can be a substantial amount of data”. Two official mechanisms keep this in check. The `--replica-cmds-style recent-actions` flag keeps only the two latest height files in `replica_cmds/`; the repository’s Docker image starts the node with it. And the repository’s Docker Compose setup runs a pruner container whose cron job, every day at 03:00, deletes files older than 48 hours from the data directory, except `visor_child_stderr`. On a host install, the same rule reads: ```bash # How much the node has written so far du -sh ~/hl/data # Same rule as the repository's pruner/scripts/prune.sh: delete files older # than 48 hours (2,880 minutes), keep the crash logs find ~/hl/data -mindepth 1 -path '*/visor_child_stderr' -prune -o -type f -mmin +2880 -exec rm {} + ``` Run it daily from cron or a systemd timer, and copy anything you want to keep to other storage first. At the README’s rate, 48 hours of retention means roughly 200 GB of logs on disk before any optional output. On a server with two drives, mounting the second one at `~/hl/data` keeps a full data disk from taking the operating system down with it. ## Ports, OS and first start - **Operating system.** “Currently only Ubuntu 24.04 is supported.” - **Gossip ports.** Ports 4001 and 4002 “must be open to the public. Otherwise, the node IP address will be deprioritized by peers in the p2p network.” Validators open 4000–4010 to other validators. - **Local endpoints.** The EVM RPC and info server listen on port 3001 on localhost; keep them private. - **Seed peers.** On mainnet, a non-validator needs at least one IP address in `~/override_gossip_config.json`. The default is 8 gossip peers; `n_gossip_peers` accepts 8 to 100. - **Signed binaries.** The `hl-visor` process verifies `hl-node` itself and will not upgrade on a verification failure, provided the public key from the repository has been imported. ```bash cd ~ echo '{"chain": "Mainnet"}' > ~/visor.json curl https://binaries.hyperliquid.xyz/Mainnet/hl-visor > ~/hl-visor && chmod a+x ~/hl-visor # pub_key.asc comes from the hyperliquid-dex/node repository gpg --import pub_key.asc curl https://binaries.hyperliquid.xyz/Mainnet/hl-visor.asc > hl-visor.asc gpg --verify hl-visor.asc hl-visor # At least one root peer is required on mainnet (replace 1.2.3.4) echo '{ "root_node_ips": [{"Ip": "1.2.3.4"}], "try_new_peers": false, "chain": "Mainnet", "reserved_peer_ips": [] }' > ~/override_gossip_config.json ~/hl-visor run-non-validator ``` Finding a peer to stream from can take a while; log lines such as `applied block X` mean the node is live. The repository’s `README_misc.md` adds a systemd unit for running `hl-visor` as a service. If you prefer Docker, note that the repository’s Dockerfile is set up for testnet by default. Our [first-hour hardening checklist](https://powerdedi.com/docs/secure-your-server) covers SSH and firewall basics. ## Which server in Tokyo PowerDedi has three configurations in [Tokyo](https://powerdedi.com/locations/tokyo), each with a 10 Gbps unmetered port, always-on network DDoS mitigation and IPMI/KVM in the browser: | Use | Configuration | Against the published specs | Price and delivery | | --- | --- | --- | --- | | Non-validator: data feed, local info server, EVM RPC | [Ryzen 9 9950X](https://powerdedi.com/dedicated-servers/amd-ryzen/ryzen-9-9950x): 16 cores / 32 threads, 128 GB DDR5, 2 × 1.92 TB NVMe | 32 threads: twice the 16 vCPUs, and the latency guide’s 32 logical cores; 128 GB, exactly the README figure; 3.84 TB of NVMe against 500 GB | $159/month; in stock, about 15 minutes | | Validator, or a non-validator with every output flag on | [EPYC 9355P](https://powerdedi.com/dedicated-servers/amd-epyc/epyc-9355p): 32 cores / 64 threads, 384 GB DDR5 ECC, 2 × 3.84 TB NVMe Gen5 | Twice the validator’s 32 vCPUs, three times its memory, 7.68 TB of NVMe | $519/month; built in 24 hours | | Several nodes, local order books, long retention | [EPYC 9555P](https://powerdedi.com/dedicated-servers/amd-epyc/epyc-9555p): 64 cores / 128 threads, 768 GB DDR5 ECC, 4 × 3.84 TB NVMe Gen5 | 128 threads and 15.36 TB of NVMe for heavy processing next to the node | $949/month; built in 24 hours | The honest caveats: - **Memory headroom.** The 9950X’s 128 GB matches the README exactly, with no margin. If you plan to run memory-hungry software beside the node, such as an order book builder or a database, take the 9355P. - **Validators need more than one machine,** as described above. The 9355P suits the validator itself; a 9950X can serve as a sentry. - **Not inside AWS.** If your goal is to sit in the same AWS availability zone as the Foundation’s peer, a dedicated server is the wrong tool: ours are in a Tokyo data center whose name we do not publish, not in AWS. Hyperliquid does not require bare metal, so a cloud instance is a legitimate choice too. For reference, Cherry Servers lists its Ryzen 9 9950X, with 128 GB, 2 × 1 TB of NVMe and a 10 Gbps uplink with 100 TB of egress included, at €439 ($513.63) a month in Tokyo, against €349 in its European and US locations: public list prices checked on October 15, 2026, with VAT not added. Our 9950X is $159 a month in Tokyo, the same as in our other cities, and traffic is not metered. Ubuntu is among the one-click images, and any other ISO can be mounted through IPMI if you need to pin an exact release. ## What running a node does not change A node is infrastructure: it gives you data, a local API and a direct line to the peer-to-peer network. It does not create an account, change who may use Hyperliquid, or alter any legal, tax or regulatory obligation where you live. Even the Foundation’s own peer comes with conditions on applicable law, sanctions and restricted jurisdictions, including certain activities involving the U.S. or Ontario. Hyperliquid describes its own latency suggestions as “not trading advice”; this guide is not trading advice either. For other chains, see our [Solana validator and RPC sizing guide](https://powerdedi.com/blog/solana-rpc-node-hardware-2026) or the [blockchain node hosting](https://powerdedi.com/solutions/blockchain-nodes) overview. ## Sources - [Running a node (README)](https://github.com/hyperliquid-dex/node) — hyperliquid-dex/node, GitHub (last updated August 10, 2026) - [README_misc.md](https://github.com/hyperliquid-dex/node/blob/main/README_misc.md) — hyperliquid-dex/node, GitHub - [pruner/scripts/prune.sh](https://github.com/hyperliquid-dex/node/blob/main/pruner/scripts/prune.sh), [pruner/cron/cron.d/prune](https://github.com/hyperliquid-dex/node/blob/main/pruner/cron/cron.d/prune) and [Dockerfile](https://github.com/hyperliquid-dex/node/blob/main/Dockerfile) — hyperliquid-dex/node, GitHub - [update spec](https://github.com/hyperliquid-dex/node/commit/f256f3190d934b8f3034b9376f83c260994a2eeb) — hyperliquid-dex/node, GitHub, May 9, 2026 - [L1 data schemas](https://hyperliquid.gitbook.io/hyperliquid-docs/for-developers/nodes/l1-data-schemas) — Hyperliquid Docs - [Foundation non-validating node](https://hyperliquid.gitbook.io/hyperliquid-docs/for-developers/nodes/foundation-non-validating-node) and [Foundation non-validating node for infra providers](https://hyperliquid.gitbook.io/hyperliquid-docs/for-developers/nodes/foundation-non-validating-node-for-infra-providers) — Hyperliquid Docs - [Optimizing latency](https://hyperliquid.gitbook.io/hyperliquid-docs/for-developers/api/optimizing-latency) — Hyperliquid Docs - [Running a validator](https://hyperliquid.gitbook.io/hyperliquid-docs/validators/running-a-validator) — Hyperliquid Docs - [Staking](https://hyperliquid.gitbook.io/hyperliquid-docs/hypercore/staking) — Hyperliquid Docs - [AWS Availability Zones](https://docs.aws.amazon.com/global-infrastructure/latest/regions/aws-availability-zones.html) and [AWS Regions](https://docs.aws.amazon.com/global-infrastructure/latest/regions/aws-regions.html) — Amazon Web Services - [CPU options for Amazon EC2 instances](https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/instance-optimize-cpu.html) — Amazon Web Services - [AMD RYZEN 9950X](https://www.cherryservers.com/pricing/dedicated-servers/amd-ryzen-9950x) — Cherry Servers (checked October 15, 2026) **Share this article** PowerDedi Engineering The team that builds, racks and runs every PowerDedi server. We write about the hardware, networks and costs we deal with every day, and we cite our sources. **Dedicated servers** $55/mo and up About 40% below the market, with a 10 Gbps unmetered port on every server. Paid in crypto. [Browse servers](https://powerdedi.com/dedicated-servers) **More from the blog** - [Guides](https://powerdedi.com/blog/ton-node-requirements) October 21, 2026 **TON node requirements in 2026: full node, liteserver, validator** - [Comparisons](https://powerdedi.com/blog/crypto-dedicated-server-providers-compared) October 15, 2026 **Dedicated servers paid in crypto: 14 providers compared (October 2026)** - [Comparisons](https://powerdedi.com/blog/hetzner-alternatives-2026) October 15, 2026 **Hetzner alternatives in 2026: crypto payment, 10 Gbps, no setup fees** - [Comparisons](https://powerdedi.com/blog/ovhcloud-alternatives-2026) October 15, 2026 **Does OVHcloud accept crypto? OVHcloud alternatives in 2026** ## Related articles [All articles](https://powerdedi.com/blog) [Guides](https://powerdedi.com/blog/category/guides) October 15, 2026 ### [TRON node requirements in 2026: FullNode, Lite FullNode and SR node](https://powerdedi.com/blog/tron-node-requirements) What a TRON FullNode, Lite FullNode or Super Representative node needs in 2026, from the official java-tron documentation: CPU, RAM, snapshot sizes, ports and JVM settings, and whether you need a node at all to send USDT on TRON. 8 min read [Guides](https://powerdedi.com/blog/category/guides) October 21, 2026 ### [TON node requirements in 2026: full node, liteserver, validator](https://powerdedi.com/blog/ton-node-requirements) What a TON full node, liteserver, archive node or validator needs in 2026, from the official docs and MyTonCtrl: CPU, RAM, NVMe and IOPS, current dump sizes, bandwidth, ports, the GRAM stake, and which servers fit. 12 min read [Guides](https://powerdedi.com/blog/category/guides) October 15, 2026 ### [Bitcoin full node hardware requirements in 2026](https://powerdedi.com/blog/bitcoin-node-hardware-requirements-2026) Disk, memory, CPU and bandwidth for a Bitcoin full node in 2026: chain size and growth, pruned versus archival nodes, Electrum server indexes, faster syncs with AssumeUTXO, and the configurations that fit. 4 min read ## Your next server is 15 minutes away. From $55/month with 10 Gbps unmetered, no setup fee and no contract. [Browse servers](https://powerdedi.com/dedicated-servers) [All articles](https://powerdedi.com/blog) Live in ~15 min · Cancel anytime · BTC, ETH, XMR, USDT or SOL --- PowerDedi — Dedicated servers with 10 Gbps unmetered, paid in crypto. Page: https://powerdedi.com/blog/hyperliquid-node-requirements. Index of the site for AI assistants: https://powerdedi.com/llms.txt