Factorio Server Hardware Requirements
"What specs do I actually need to run a 24/7 Factorio server?" is one of the most common questions from people moving off a player-hosted game onto a dedicated box. The short answer surprises most newcomers: a Factorio dedicated server is very light on hardware compared to most games, and the spec that matters is not the one most people shop for. This guide breaks down what the headless server really needs for CPU, RAM, storage, and network, and explains why single-thread CPU speed and UPS decide performance more than core count or player count.
The one-line summary: Factorio simulates almost the whole factory on a single thread. A fast CPU core beats many slow cores every time, RAM scales with how big your factory gets (not how many cores you have), and adding players costs almost nothing compared to adding more factory.
CPU: Single-Thread Clock Speed Is King
Factorio's game simulation is overwhelmingly single-threaded. Wube have parallelised specific subsystems over the years and their own changelog dates each one: electric networks, fluids and heat pipes in 0.18.0 on 21 January 2020, belt update logic in 1.1.0 on 23 November 2020, and the circuit network in 2.0.7 on 20 October 2024. Everything else that advances the world, including inserters, assemblers, trains and the entity update loop itself, runs on one thread, and the tick is deterministic (that determinism is also why every client must stay in sync). Those three carve-outs are specific subsystems, not a general move to many cores, so the practical consequence has not changed:
The --threads launch argument is not a counter-example, and it is the one people cite. The option parser's own description of it is "Number of threads to use when generating map previews". It sets the thread count for rendering preview images, not for the simulation, and passing a large number to it will not make your factory run faster.
- More cores do not speed up your factory. A 16-core server is no faster at simulating one base than a 4-core server with the same per-core clock.
- A high single-thread clock does. A modern CPU with strong single-core performance lets a larger factory run before the simulation falls behind.
- You only really need one fast core for the simulation itself, plus a little headroom for the operating system, autosave compression, and network handling.
UPS: The Metric That Actually Measures Server Health
Factorio targets 60 updates per second (UPS), which is one simulation tick every 16.67 milliseconds. As long as the server holds 60 UPS, the game runs at full real-time speed for everyone. When the factory grows complex enough that the CPU cannot finish a tick in 16.67 ms, UPS drops below 60 and the whole game slows down for all connected players at once, because everyone is locked to the same simulation.
This is the number to watch, not CPU percentage. A server can show modest CPU use and still drop UPS if a single tick occasionally overruns. The official Tutorial: Diagnosing performance issues on the Factorio Wiki explains how to read the in-game timing overlay (press the debug key and enable show-time-usage) to find which part of the tick is the bottleneck.
Stop guessing: the headless build benchmarks itself
You do not have to argue about single-thread scores, because the same binary that runs your server will measure it. The option parser ships a benchmark mode, and these are its arguments with the parser's own descriptions:
| Argument | What it does |
|---|---|
--benchmark FILE | load save and run benchmark |
--benchmark-ticks N | number of ticks for benchmarking. Default is 1000 |
--benchmark-runs N | how often the number of ticks will be run, map will reload after each run. Default is 1 |
--benchmark-verbose timings | comma separated list of timings to output each tick; "all", "timestamp" and every other value visible in the debug view are allowed |
--benchmark-sanitize | only output the final benchmark results |
--benchmark-ignore-paused | leaves the game paused if it was paused when saved; by default the game is unpaused when a benchmark starts |
--until-tick TICK | run a save until a given map tick |
That turns the whole "will this box hold my base" question into a measurement. Take the save you actually play, run it on the candidate hardware with a few thousand ticks and several runs, and compare the average tick time against the 16.67 ms budget. A save that benchmarks at 8 ms a tick has roughly half its headroom left; one at 17 ms is already below 60 UPS and will feel it the moment somebody builds something. Because the same save on the same build is deterministic, this is a fair comparison between two machines in a way that a synthetic CPU score is not.
Two practical notes. --benchmark-runs reloads the map between runs, so it measures steady state rather than one lucky pass, and it is the flag to use when a result looks too good. And --benchmark-verbose all breaks the tick down by subsystem, which is how you find out whether your problem is belts, fluids, trains or your own circuit network before spending money on a CPU that will not fix it.
RAM: Scales With Factory Size, Not Player Count
A headless server does not load graphics, sounds, or sprites, so it needs far less memory than a full game client. Rough working figures, consistent with long-running threads on the Factorio Forums:
| Factory stage | Practical RAM |
|---|---|
| Bare minimum to boot a tiny map | ~1 GB (technically runs, fine for a small early-game world) |
| Small group, early-to-mid game | 2 GB comfortably covers most factories |
| Large base or Space Age multi-planet save | 4 GB and up |
| Sprawling megabase | 4-8 GB, mostly driven by entity count and map size |
What does not drive memory much is the number of connected players. Adding a friend to the server costs a small amount of network state, not the gigabytes that adding another thousand assemblers would. If your server feels slow, the cause is almost always CPU (a dropping UPS), not RAM. RAM exhaustion shows up as crashes or refusal to load a large save, not as a gradual slowdown.
Storage and Network
- Disk: The game install plus a save folder is small (a few GB total). What matters is write speed during autosaves. On a busy server, autosave can briefly stutter the simulation while the save file is written, so an SSD or NVMe drive keeps that pause short. The shipped settings autosave every 10 minutes into 5 rotating slots, so budget five copies of your save, and keep the slots so a bad save never costs you the world. There is also a
non_blocking_savingsetting, which the settings file itself labels highly experimental and warns you to enable only at your own risk of losing saves: on UNIX systems the server forks itself to make the autosave, so plan for some memory headroom rather than sizing RAM to the exact working set. - Network: Factorio multiplayer is bandwidth-light because only player inputs travel over the wire, not the full game state. A handful of players use a few hundred kbps. What players feel is latency and packet loss, so a stable connection matters more than raw throughput. The default port is UDP 34197. The settings file puts the network tick rate at 60 packets a second by default and states its own bounds, a minimum of 6 and a maximum of 240, and caps map uploads to joining players at 5 concurrent slots. That upload slot count, not bandwidth, is usually what makes six people joining at once after a restart feel slow.
"Is a Cheap Mini PC Enough for a 24/7 Server?"
For a small group on an early-to-mid game factory, yes. A low-cost mini PC or a small VPS with one fast core and 2 to 4 GB of RAM will happily host a 24/7 dedicated server for friends. The questions to ask before buying one are about single-core performance, not core count or total RAM:
- What is the CPU's single-thread benchmark score? Higher is better, and it directly sets the megabase ceiling.
- Is there a quiet, low-power thermal envelope that can run continuously without throttling? Sustained clock matters more than a high turbo boost it cannot hold.
- Is storage an SSD rather than an old mechanical drive? This keeps autosave pauses short.
A Steam Deck, an old laptop, or a Raspberry-class board can also run the headless server, with the caveat that low-clock ARM and mobile chips will hit the UPS wall on a smaller factory than a modern desktop core. They are fine for a casual save and a poor choice for a megabase.
Quick Sizing Cheat Sheet
| Use case | CPU | RAM |
|---|---|---|
| 2-4 friends, vanilla early/mid game | 1 fast modern core | 2 GB |
| Larger group or moderate mods / Space Age | 1-2 fast cores | 4 GB |
| Megabase, heavy circuit logic, big train networks | The fastest single-core clock you can get | 4-8 GB |
Related Reading
- How to host a Factorio headless server
- Auto-pause and persistent 24/7 factories
- Fixing multiplayer desyncs
- Space platform and UPS optimization
- Official: Factorio Wiki - Diagnosing performance issues
Skip the Hardware Math with Supercraft
Supercraft runs Factorio dedicated servers on nodes chosen for high single-thread clock speed, the spec that actually moves UPS, with NVMe storage for fast autosaves and tuned autosave slots. You pick a plan and play, no mini-PC sourcing, no thermal throttling, no port forwarding.