Palworld Server Setup and 64-Player Configuration Checklist

A Palworld server that runs perfectly for six friends can fall over within twenty minutes of being listed publicly, and the cause is almost always one of three resource ceilings being crossed at the same time. This guide covers the setup checks that prevent it, organised by the failure mode each one addresses rather than by menu order.

It assumes a bare-metal or dedicated-vCPU machine. The community-server panel model — one container per world, shared CPU, no shell — is convenient for a private group but removes the tools you need when 64 players show up at once. If you are still picking hardware rather than configuring it, size the machine against the Palworld hardware sizing table first and return here for configuration.

Before First Boot: OS and Runtime Choices

Install a current LTS Linux distribution with a recent glibc — Debian 12, Ubuntu 22.04 or later. The Palworld server binary ships as a Windows executable and via SteamCMD for Linux, running under Proton or the native Linux build depending on the version. Use SteamCMD for installation, keep the install directory on the same filesystem you will back up, and run the process as a dedicated non-root user.

The community-server container images are a reasonable starting point, but strip anything you do not need: disable the bundled RCON web panel if it is exposed, and make sure the container gets a real memory limit rather than inheriting the host’s. A container with a 512 MB limit will be OOM-killed during world save and you will only find out from a player complaint.

The Settings File: What to Change and What to Leave

Configuration lives in PalWorldSettings.ini under Pal/Saved/Config/LinuxServer/. Every value sits on a single OptionSettings=(...) line, so a malformed edit breaks the server with no useful error. Back up the file before editing, and validate by starting the server with the new file before opening the ports to players.

SettingDefaultPublic server valueWhy
ServerPlayerMaxNum32Match hardware, not habitOversubscribing slots is the #1 cause of tick collapse
AutoSaveSpan3060–120Halves save-write hitches
DropItemMaxNum30001500Reduces replicated entities
BaseCampWorkerMaxNum1515–20Largest single RAM driver per base
PublicPort / RCONPort8211 / 25575Change RCON, restrict by firewallRCON is unauthenticated by default in some builds
ServerPasswordemptyRequired for private worldsOpen servers get scraped and griefed within hours

Two defaults are frequently misread. bIsMultiplay must be true or the server runs in single-player mode with no network listener, and Difficulty changes raid scaling rather than load. Neither affects performance, but both cause “server not visible” tickets more often than any networking issue.

Networking: Ports, Firewall, and Discovery

Palworld needs UDP 8211 by default, plus the query port if you use one, plus TCP for RCON if enabled. Open only those. A common failure at large player counts is conntrack table exhaustion on the host when the firewall tracks every UDP flow; raise net.netfilter.nf_conntrack_max to 262144 and check nf_conntrack_count under load.

For discovery, the server list relies on a beacon rather than a central registry. If your node sits behind a NAT or a filtered uplink, players will not see the server in the in-game browser even though direct connect works. Publishing the address to your community Discord is more reliable than debugging the beacon, and it removes an external dependency at launch.

Setup Checklist for a Stable First Week

  • Dedicated non-root user, systemd unit with Restart=on-failure and a restart delay of 30s
  • Memory limit or cgroup for the server process, with an OOM handler that alerts you
  • World directory on local NVMe, backed up every hour to a separate disk and daily off-host
  • RCON bound to localhost or restricted by source IP
  • Server password for launch, removed deliberately once you have moderation in place
  • Log rotation configured — Palworld logs grow quickly at 32+ players
  • Uptime and player-count monitoring with an alert on process exit
  • A documented restart procedure that copies the save first

The systemd unit matters more than most guides suggest. Without Restart=on-failure with a delay, a crashing server can enter a boot loop that corrupts the world save because it restarts mid-write. Set StartLimitBurst conservatively so the unit stops trying after five failures in ten minutes and leaves the save intact for inspection.

Diagnosing the Four Common Failures

Server invisible in list: port not open, NAT not forwarded, or beacon blocked. Verify with ss -ulpn | grep 8211 and an external UDP probe before touching config.

Rubber-banding at peak: tick thread saturated or save writes blocking the main loop. Check per-thread CPU and disk latency rather than raising the tick rate.

Random disconnects for everyone simultaneously: memory pressure causing swap, or a provider-level network blip. Correlate timestamps with vmstat and mtr.

World fails to load after restart: interrupted save. Restore from the pre-restart backup; do not attempt to repair the file in place on the only copy. The full failure taxonomy and triage order is in the crash-loop triage guide, which applies to Palworld despite being written around modded servers.

Storage, Saves, and Growth Over Time

World size grows faster than most operators expect. A 40-hour world is typically 150–250 MB; a 500-hour community world with extensive bases reaches 1–2 GB. Saves are written as a full copy with a backup rotation, so peak disk usage during a save is roughly twice the world size. Keep at least 20 GB free beyond current usage.

Set bIsUseBackupSaveData=true and rotate copies yourself on top of it, because the in-game rotation can overwrite a good backup with a corrupt one when saves happen close together. An hourly snapshot with 24-hour retention covers almost every realistic recovery scenario for a community server.

When to Move Off a Hosted Panel

Three signals indicate it is time: sustained tick FPS below 25 with under 30 players, repeated OOM kills you cannot prevent through config, or a need for IP-level firewall and kernel tuning the panel does not expose. At that point a dedicated machine with IPMI access gives you the ability to pin the tick thread, disable CPU power states, and choose storage — levers that are unavailable inside a managed container. Current configurations and the hardware each maps to are listed on our dedicated server plans page.

Migration is straightforward if you prepared for it: stop the server, copy the entire saved directory, start on the new machine, and verify the world loads before announcing the new address. Keeping the old server running read-only for 24 hours gives players time to update their bookmarks without losing progress.

Leave a Reply