Game-server hosting benchmarks, methods and buying guidance
Game server hosting, without the made-up benchmarks
GameServerBench is a game-server hosting site that reads what hosts publish (prices, plan limits, refund rules, testing terms) and what other publishers report, then shows who said what. As of 8 November 2026 it has measured nothing: there is no test rig, no rented test server and no tick, memory or latency result anywhere on the site. Plenty of hosting pages say they tested; this one says it has not, and says what it would record if it did.
CH A
No input
––– ms/div
No run recorded
Stable players
–––players
Tick time p95
–––ms
Peak memory
–––GB
Join time
–––s
Method first
A result is only as good as the path it took to get here.
These pages set out what a number has to carry before it can appear here. Read them before you trust a chart on this site.
- 1Test rigWhat the measurement runs on, and how it is recorded.
- 2Load profilesWhat the server is asked to do during a run.
- 3Stable playerThe threshold a server has to hold to count a player.
- 4IndependenceWho pays for what, and what host terms allow.
A number appears here only with all six of these attached.
- Method version
- Run date
- Host and plan
- Game and version
- Load profile
- Raw log
Benchmarks by game and pack
Published evidence to read today
- Palworld3 guides · 9 hosts list it
- Valheim2 guides · 9 hosts list it
- Project Zomboid2 guides · 7 hosts list it
- All the Mods 102 guides
- Enshrouded1 guide · 6 hosts list it
- Hytale1 guide · 6 hosts list it
- SatisfactoryFramework only
- RustFramework only
- 7 Days to DieFramework only
- CobblemonFramework only
- Modpack hostingFramework only
- Latency by regionFramework only
Host files
Alphabetical. No host is ranked until a run exists.
Tools
Cost arithmetic and published facts. No measured data.
Where should you start?
Pick the situation closest to yours.
- Renting a server for friends. Choose the game on the benchmarks hub, for example Palworld or Valheim, to see which hosts list it, at what published price and refund window.
- A modpack that lags. Learn to read MSPT and TPS, then follow the profile-reading steps in the ATM10 lag guide. It applies general fixes and reports no measured gains.
- Deciding whether to self-host. Work through the Palworld or Valheim cost comparison, or enter your own numbers in the rent vs self-host calculator.
- About to pay a specific host. Open its review on the hosts hub first. Every review covers the price after the first month, the refund rule, the stated hardware and the host's testing terms, or says plainly that they could not be read.
- Wanting to test your own server. Use the guide to benchmarking your own server, which starts with checking your host's terms.
For anything else, the guides hub is organised by question: how much RAM, why it lags, rent or self-host, how to read numbers, what it costs.
What is worth reading first?
Four things on the site are not found in the usual "best host" lists.
- The host terms comparison. Independence, host testing terms and conflicts sets out what each host's terms say about benchmarking, bots and heavy use, and what GameServerBench's own conflicts of interest are. None of the terms read grants a right to benchmark.
- Refund fine print in the reviews. Refund windows are short and the details differ. G-Portal states its refund rule in sentences that disagree about when the clock starts, Sparked Host gives three windows on three pages, and Shockbyte allows 72 hours, once per client.
- Who-says-what requirement tables. RAM advice conflicts. The Palworld and ATM10 guides list each figure with its publisher and whether that publisher sells hosting. The cheapest hosting guide shows first-month against renewal arithmetic.
- The how-to-benchmark guide. It tells you how to measure fairly yourself, including why idle bots mislead on modded servers.
How does the site stay honest?
Host claims are written as claims: "Host Havoc lists", not "Host Havoc has". Third-party figures carry the publisher, the date and whether it sells hosting. Where a page could not be read, the page says so. No host is ranked, and no score is given. Some pages may earn a commission later; no affiliate links are live, and the disclosure explains the rules. About covers who is behind the site.
Measured data is the plan, not the present. The format is described on the data page and the two planned recurring reports on the reports page. Both say plainly that they are empty.