How to compare Satisfactory server hosts

Why does a late-game Satisfactory factory change the choice?

A late-game save is the load event for Satisfactory, and the wiki ties hardware to save size as well as player count. The Official Satisfactory Wiki (last edited 4 November 2026) says of its figures that the server "may or may not run with less powerful hardware, depending on player count and save file size". No hardware table from Coffee Stain itself was found.

The same page quotes a developer from November 2021: the server "spawns up to 26 threads" but "the game loop is still fundamentally single threaded". That is why the wiki asks for a single-thread rating of 2000 or higher, and warns that on a virtual machine "a kvm64 CPU won't work".

Saves add a second cost. The configuration page (20 May 2026) gives a default autosave every 300 seconds and says shorter intervals mean "a drop in server performance while the server is saving". The main wiki page adds that raising Network Quality "may improve client load times and network performance at the cost of server framerate", so a host's defaults matter.

Whose Satisfactory figures conflict, and who is selling?

Published guidance disagrees on RAM and, more sharply, on player count.

Two conflicts matter. The wiki's configuration page says the server "is tuned to be most performant with the default player cap of four" and that raising it "may have negative effects", while Host Havoc sells up to 28 slots. And WinterNode presents the wiki's numbers as "Coffee Stain's own documentation", while the wiki calls them community recommendations.

How are Satisfactory plans sold?

Satisfactory plans are sold by player count or slot, which sets who can join and says little about factory size. Two hosts were confirmed to sell Satisfactory on pages read 8 November 2026, listed in no order of merit. Other hosts may also sell it; these are the ones whose Satisfactory pages were opened.

Uncapped has a limit worth reading. Indifferent Broccoli's terms (28 October 2026) cover a server using "far more memory than its game and player count call for", and the host may restart it or limit its memory. A two-player megafactory is the case to ask about.

Which fits, on published plan structure only: Indifferent Broccoli prices a group of 1 to 4 players at $5.99 to $9.99 a month and states uncapped RAM subject to fair use, while Host Havoc starts at 4 slots for $15.00 a month and shows no RAM per plan. Neither host's performance with a large factory is known. The Indifferent Broccoli and Host Havoc reviews cover wider plans.

What should you ask a host before moving a factory?

Ask these before paying, because the plan pages read do not answer them.

• Which CPU, and what is its single-thread rating? Compare it with the wiki's 2000 floor, and ask whether a VM exposes a kvm64 CPU type.
• How is a large factory with few players treated? Get the answer in writing on any uncapped or slot plan.
• How long are backups kept? The server rotates three autosaves by default, per the configuration page, which is about 15 minutes of history at one save every 300 seconds (3 × 5 minutes).
• Can you change the settings that matter? Autosave interval, Network Quality, player cap and game branch.
• Can you reach the server's API?

How can you judge a plan using your own save?

The most useful Satisfactory test is your own factory, read through the server's own reporting, during a trial or refund window and under that host's terms. The wiki documents that saves upload through Manage Saves in the Server Manager.

The HTTPS API page (20 May 2026) lists an AverageTickRate field in QueryServerState, in ticks per second, and a HealthCheck that returns "healthy" above ten ticks per second. Most API functions need an authentication token, and the API shares TCP port 7777, so ask whether your panel lets you reach it.

1. Note AverageTickRate after the save loads.
2. Read it during ordinary play across busy hours.
3. Note how long the game stalls at each autosave.
4. Compare with the configured 30-tick default. The wiki says that "if the hardware can't hit the default 30, increasing it won't help".

One reading on one evening is a start; shared hosts vary by the hour. How to benchmark your own server lists further readings. Check the terms first and see independence and host terms, since silence is not permission and some terms bar software beyond the game itself.

What would be measured, and why?

The planned protocol would load one late-game save onto three hosts and log the following. It would not use player bots, and whether any load client is acceptable on a given host is unresolved.

One save is one factory design, so one run cannot be generalised; see load profiles and MSPT and TPS explained. No test server exists, and a late-game save must be built or sourced with the right to describe it. Next: the benchmarks hub or the method overview.