How to choose game server hosting: a checklist

Most hosting comparison advice is about features nobody uses. The decisions that actually matter are four: whether the route between your players and the machine is good, what the CPU share really is and what happens when you hit the memory limit, whether the backups are yours or the host's, and how quickly you could leave. Everything else is preference.

You can check all four before you spend anything, and three of them take ten minutes. The fourth - support - you test by asking a question and timing the answer. This post is a checklist you can use on any host, including ones we compete with, and it is deliberately written so that the answers to some of these questions make us look worse than a bigger company would. The point is that you should go and ask.

Test the latency yourself, first#

Location is the only thing hosting genuinely cannot change after you buy, so measure it before anything else.

Ask the host for a test IP address on the node or network you would be on. A host that has one ready is a host that expects to be measured. Then, from the connection you will actually play on, at the hour you will actually play:

$ ping -c 50 203.0.113.10$ mtr -rwzbc 100 203.0.113.10

On Windows, ping -n 50 and tracert, or install WinMTR. What you are looking for:

• Mean round trip. Under 30 ms is excellent, 30-60 ms is fine for nearly every game, 60-100 ms is noticeable in shooters and irrelevant in survival games, above 120 ms is a choice you should make deliberately.
• Jitter, the variation between pings. A steady 70 ms plays better than a 40 ms average that swings between 20 and 140. Games buffer for latency and cannot buffer for unpredictability.
• Loss at the final hop. Loss in the middle of an mtr that does not carry through to the end is routers deprioritising ICMP, not a problem. Loss that starts at a hop and persists to the destination is a real problem on that path.

There is a floor you cannot buy your way under. Light in fibre covers roughly 200,000 km per second, so a round trip costs about 1 ms per 100 km of cable, and cable does not run in straight lines. London to Frankfurt is about 640 km, so anything under 10 ms is physically impossible and 12-18 ms is a good real-world result. If a host advertises a number below the speed of light for your distance, they are advertising a different measurement.

Two things not to do. Do not trust IP geolocation databases to tell you where a server is; they are frequently wrong by a country and are updated by the host. And do not test only from your own connection - if half your players are on a different continent, test from one of theirs. Choosing where your server lives is about picking for the group rather than for the loudest person in it, and latency, jitter and packet loss is what to send a player who says the server is laggy.

Read the specification: RAM, CPU share and disk#

Game hosting is sold by memory because memory is the number people understand. It is rarely the number that decides whether the server runs well.

Ask what the CPU allocation is, in a unit. A good answer is a percentage of a core or a vCPU figure - "200% CPU" means two cores' worth, and it should say whether that is a hard limit or a burst. A bad answer is "unlimited CPU", which on a shared node means "whatever is left after everyone else", and a worse answer is no figure at all. Most game servers are limited by single-thread performance, so also ask which CPU generation the nodes run; a 2016 Xeon with a 2.1 GHz base clock and a 2024 desktop chip are not the same product at the same "2 vCPU". CPU or RAM for game servers is thirty seconds of diagnosis that tells you which one you are buying for.

Ask what happens at the memory limit. There are two designs and they feel completely different. Either the container is stopped and restarted clean, which costs whoever was online two minutes and produces a clear event you can see; or the process is allowed to swap, which produces a server that is technically online and unplayable for half an hour. Neither is wrong, but you should know which one you have bought, because it changes how you read a sudden restart. For the record, RE:NODE does the first: at the limit the kernel stops the container and it restarts clean rather than swapping, which is the opposite of upstream Pterodactyl's default.

Ask whether disk is a quota, and what kind. NVMe, SATA SSD and spinning disks are three different experiences for a game that saves a world under load. Ask whether the quota is enforced, and what happens when you reach it - a full disk corrupts world saves mid-write. What NVMe actually changes is specific about which operations improve and which do not.

Then do the arithmetic that hosts do not publish. Work out roughly what a node costs: a machine with 128 GB of RAM, NVMe and a decent CPU, in a European facility with bandwidth, is a few hundred a month. If a host sells 10 GB plans at $2, they need to fit an implausible number of them on that machine to cover it, and the way that is achieved is by hoping most of them are idle. That hope is called overselling and it is the mechanism behind shared CPU and noisy neighbours. Cheap is fine. Impossible is a warning.

What the panel actually lets you do#

Most game hosts run Pterodactyl or something derived from it, which is good news - it means the capabilities are broadly known and you can find out what you are getting. Pterodactyl explained covers the architecture. What varies between hosts is how much of it they leave switched on.

Check for these specifically, because their absence is how a cheap plan becomes an expensive afternoon:

The single best test is to ask whether you can upload and run a build of the server software that the host does not list. If the answer is no, you are renting a preset rather than a server, and the day your modpack needs a specific loader version you will be opening a ticket instead of a file manager.

Ask about the database too. Game servers that need one - economy plugins, roleplay frameworks, statistics - need a database slot with a host, user and password you can see. Ask how many are included and whether you get a real client connection or only a web tool.

Backups, and whether they are really yours#

This is where cheap hosting is most often cheap.

• Are backups included, and how many? Not "we back up the nodes". Node-level backups protect the host from hardware failure. They do not help you when you delete the wrong world folder, because you cannot ask for one file from them.
• Can you take one yourself, right now, without a ticket? If not, you do not have backups; you have a hope.
• Can you download one? A backup you cannot download is a backup that only exists as long as your relationship with the host does. This is the single most important question on this page.
• Where are they stored? On the same machine as the server is a copy, not a backup. Off the machine is a backup.
• What happens when you cancel or when the server is deleted? On most panels, including ours, deleting a server deletes its backups with it - locked ones included. Find out before you cancel anything.
• Is there a retention period? Rotation that silently removes your oldest copy is fine as long as you know the number.

Then do the thing almost nobody does: restore one in your first week, while there is nothing to lose. A backup nobody has restored is a hypothesis. Backups that actually restore and testing a restore before you need it are about that half of it, which is the half that fails.

Ports, mods and what a host quietly forbids#

Two policy questions decide whether a plan fits your plans.

Ports. Ask how many allocations the plan includes and whether you can add more yourself. Some games need three or five: game, query, RCON, and a voice or web-admin port on top. Arma 3 wants several. A plan with one allocation is fine for Minecraft and wrong for DayZ. Ask whether query and RCON ports count against the total, and whether adding a port is self-service or a ticket. Game server ports explained covers why games need more than one.

Mods. Ask whether there is a list of allowed mods or plugins. Some hosts maintain one and it is the reason a modpack will not install. Ask whether you can upload a custom server binary, and whether custom container images are possible if you need something unusual. RE:NODE has no whitelist of allowed mods - you upload what you like through the file manager or SFTP - and does not offer custom images, which is a real limit worth knowing.

Suspension policy. Ask what gets a server suspended. Sustained high CPU should not: a server sitting at 100 per cent of its share is slow, not broken. Repeated crash loops often do, and there is usually an automatic system behind it. Ours warns after three unexpected restarts in an hour and suspends at six, which is a policy you would rather read here than discover at midnight - why your game server keeps restarting is the thing to fix first.

Credentials you must supply. Several games need something from you before they will run: a Steam game server token for CS2 and Unturned, a Cfx.re key for FiveM, a Klei token for Don't Starve Together, an auth key for BeamMP, and a full Steam login for Arma 3, DayZ, Project Zomboid and 7 Days to Die. That last category means handing a password to a panel, so use a spare Steam account for it. Ask any host how those are stored. Steam game server tokens explains what the token is and how it gets banned.

DDoS, uptime and how to read a claim#

Two numbers appear on every hosting page and neither means what it looks like.

"DDoS protection, up to N Gbps." The capacity figure is the least useful part. The questions that matter are: is filtering always on or triggered after an attack starts, in which case the first minute is an outage; does it handle UDP reflection and malformed packets, which is what most game server attacks are; what happens to an attack that looks like legitimate player traffic; and - the important one - do you null-route the address under attack? Null-routing means dropping all traffic to your IP until the attack stops, which is cheaper for the host and identical to an outage for you. Many budget hosts do it and none advertise it. Ask directly.

Be suspicious of anyone who promises no downtime under any attack. A host that promises that is selling you a sentence, not a service. The honest version is narrow: upstream filtering drops obvious volumetric floods, reflection and malformed traffic, and attacks shaped like real players are not filtered anywhere, by anyone. What filtering can and cannot do is the longer version.

"99.9% uptime." Work out what it permits: 99.9 per cent is about 43 minutes of downtime a month, 99.5 per cent is 3.6 hours, and 99.99 per cent is 4.3 minutes. Then ask the question behind it - is there an SLA document, what does it pay, and who has to claim? Most consumer hosting SLAs pay in service credits you must request within a window, which is worth approximately nothing on a hobby server. A published uptime figure with no document behind it is marketing. So is no figure at all: RE:NODE samples availability every five minutes and publishes no percentage, which you should weigh as the absence of a promise rather than as modesty.

Billing, refunds and the exit test#

The money questions are boring and they are where people get stuck.

• Terms. Monthly, and usually cheaper per month on 3, 6 or 12. Do not commit to a year on a host you have not run for a month.
• Renewal. Is it automatic? Do you get a notice, and how long before? Ours emails a renewal invoice 7 days before expiry and again on the day, which is the behaviour to look for.
• Non-payment. What is the grace period, when is the server suspended, and when is the data actually deleted? This number is the one that costs people worlds.
• Refunds. Is there a window, and is it self-service or a ticket? RE:NODE has no automatic refund button - refunds go through a ticket - and no free tier, so there is no way to try before paying. That is a genuine disadvantage against hosts that offer a money-back window, and you should weigh it.
• Currency. If you are billed in a currency you do not hold, check whether conversion happens at your bank or theirs, and expect the number to move.
• Delivery time. Game, app, web and database servers here are created when payment clears, usually inside a minute; a VDS is prepared by hand within 24 hours. Ask, because "instant setup" often is not.

Then the exit test, which is the single best way to judge a host you already use. Within your first week, download a full backup and a copy of your files. If you can do that in an hour without a ticket, you can leave whenever you want, and everything else is negotiable. If you cannot, you are locked in regardless of what the terms say. Moving a server without losing players is the procedure when you decide to go.

Red flags worth walking away from, in rough order of how much they predict:

• Lifetime plans. Hosting has recurring costs. A one-off payment for forever means the money runs out and so does the server.
• "Unlimited" RAM, slots or CPU. Every one of these is metered by physics. Unlimited means undisclosed.
• No CPU figure anywhere on the site.
• No company details, no address, no legal entity. You are trusting someone with data and money.
• Support only through Discord direct messages. No record, no ticket, no accountability, and no way to reach anyone when the person is asleep.
• Reviews all posted in the same fortnight.

What to send in a pre-sales ticket#

Copy these into a pre-sales ticket. The answers matter, and so does how long they take and whether they are answered like a person or a template.

1. What is the CPU allocation on this plan, in vCPU or percentage of a core, and is it a hard limit?
2. What happens when a server reaches its memory limit?
3. Can I have a test IP on the node I would be placed on?
4. How many backup slots does this plan include, are they stored off the machine, and can I download one?
5. What happens to my backups if I cancel or the server is deleted?
6. How many port allocations do I get, and can I add more myself?
7. Is there a list of allowed mods or plugins, and can I upload my own server build?
8. Is DDoS filtering always on, and do you ever null-route an address under attack?
9. What is the refund policy, and is there a trial?
10. How long after non-payment is a server's data deleted?

A host that answers all ten in plain language is telling you a great deal about the next twelve months. A host that answers six and deflects on backups and null-routing is telling you something too.

FAQ#

How much should a game server cost?

Enough to pay for the hardware it runs on. Most small game plans sit between $3 and $10 a month depending on the game's memory floor, and the biggest driver of the price is RAM. How much a game server costs breaks down what you are actually paying for.

Does the host's location matter more than the specification?

For latency, yes, and it is the only thing you cannot change later. Pick the location for where your players are, then pick the specification. A faster machine further away is usually the worse server.

Is a cheap host always a bad host?

No - small operators with low overheads are genuinely cheaper than large ones. What is always bad is a price that cannot cover the hardware, because the only way to make that work is to put too many servers on a machine and rely on most of them being idle.

Should I buy a year up front for the discount?

Not on a host you have not run for at least a month. Longer terms are genuinely cheaper per month, but they are only cheaper if you stay, and the discount is usually smaller than the cost of being stuck.

What is the difference between a game panel and a VDS?

A panel gives you one pre-configured server and does the operating system work for you. A VDS gives you root and the whole job. Choosing between a VDS and a game panel has the comparison, and running several servers on one VDS is what the VDS route looks like in practice.

How do I test a host without committing?

Ask for a test IP and measure the route, read the panel documentation, and send the ten questions above. If there is a refund window, use the first few days of it to restore a backup and download your files rather than to admire the dashboard.