Hardware and method

What your server actually runs on

Most hosting pages describe hardware in adjectives. This one gives you the specification we read off the machine, how we back it up and watch it, how we measure the numbers we publish — and the things we do not have yet.

Compare plans and pricing Ask us what fits

Written and reviewed by YiorasHost technical team · Last tested

The node

Read from the machine on 25 July 2026

Java Edition servers run on one production node. This is what it reports about itself, not what a datasheet says it should be.

Provider and type Hetzner Cloud instance — virtualised, not dedicated hardware
Location Helsinki, Finland — availability zone hel1-dc2, read from the instance metadata
Processor 4 vCPU, AMD EPYC (Rome generation)
Memory 8 GB total, of which 6 GB is allocatable to customer servers
Storage ~76 GB, non-rotational with TRIM support; 60 GB allocatable to servers
Operating system Ubuntu 24.04 LTS
Networking IPv4 and IPv6, with Hetzner network-level L3/L4 DDoS mitigation
About the word NVMe. We say NVMe elsewhere on this site because that is Hetzner's published specification for Cloud servers, and from inside the instance we can confirm the disk is non-rotational and supports TRIM. We cannot see the physical device through the virtualisation layer. So treat NVMe as a claim we are passing on from our provider, not one we measured — that distinction matters more to us than the word does.
What you are buying

Dedicated memory, shared everything else

The RAM in your plan is reserved for your server. It is not shared, and nothing else can take it.

CPU time, storage throughput and network capacity are shared with the other servers on the node. That is normal for hosting at this price, and it is the honest description of it. A host charging these prices while implying dedicated cores is describing something else.

Each plan also carries a CPU limit, so a single server cannot monopolise the node. It is a ceiling on what you can take, not a floor guaranteeing what you always get.

Backups and monitoring

What happens while nobody is looking

Both of these were verified on 25 July 2026, the day this page was written.

Offsite backups

  • Server data is backed up with Borg to a Hetzner Storage Box, which is separate storage from the machines it protects.
  • The game node backs up nightly; the billing and panel host every six hours.
  • Both most recent runs completed successfully — the node keeps 8 archives and the control host 14.
  • We have restored from these backups as a test rather than assuming they work, because an untested backup is a guess.
  • You can also take and download your own backups from the panel whenever you want.

Monitoring and alerting

  • Prometheus collects metrics, Grafana displays them, Alertmanager raises alerts and Uptime Kuma checks the public services.
  • It runs away from the production hosts on purpose, so a failure of the thing being watched does not take the watcher with it.
  • Alerts reach us by email, and we have tested that path rather than trusting the configuration.
  • We track memory and CPU committed versus actually used, which is what tells us when to add capacity before you notice.
How we measure

The rules our benchmarks follow

Any number on this site should be checkable. These are the rules we hold ourselves to, and you should hold us to them too.

Named conditions

Every figure states the software version and build, the Minecraft version, the world seed and the settings used.

Repeated, not cherry-picked

Runs are repeated and reported as a median with the full range, never as the best result we saw.

Measured on a test machine

Benchmarks run on an internal fixture, not the production node. Every page carrying a number says so beside the number.

Retracted when wrong

When we publish something incorrect we correct it and record the retraction, rather than quietly editing the page.

A worked example

Our startup measurements compare Paper, Fabric, Forge and NeoForge on one machine, on Minecraft 26.2, with the world seed pinned so every run loads an identically-seeded world, three consecutive restarts each.

That method exposed something worth knowing: the startup time a Minecraft server prints in its own console is not comparable between server software. Three of the four report around a quarter of a second where Paper reports over eight, because each starts its timer at a different point. Measured properly, from the first console line to the ready line, all four land within about two seconds of each other.

We publish both numbers and tell readers to ignore the misleading one. A host optimising for impressive figures would publish only the flattering column.

See the four-way comparison → Paper measurements → Network latency measurements →
The uncomfortable part

What we do not have

You can find this out after ordering, or you can read it here first. We would rather you read it here.

One node, one location

Every Java server runs on a single node in Helsinki. There is no second machine to fail over to, and no second region. If that node goes down, servers on it go down with it.

No uptime guarantee

We do not publish an SLA percentage, because with one node we cannot honestly back one. Monitoring and alerting are real; a promised number would not be.

Virtualised, shared resources

This is a cloud instance with shared CPU and I/O, not dedicated hardware. Heavy load elsewhere on the node can affect you.

No tick-rate figures

We have not measured TPS with a method we would defend, so we do not publish TPS claims. Startup time is not a substitute, and we say so on the pages that measure it.

Limited capacity right now

Larger plans are out of stock until we add a node. We would rather show that than sell you a server we cannot build.

If any of that rules us out for what you are building, it is better for both of us that you know now. Ask before you order.

FAQ

Infrastructure questions

Is my server on dedicated hardware?

No. Servers run on a Hetzner Cloud instance in Helsinki, which is virtualised. Your plan’s RAM is reserved for your server, but CPU time, storage throughput and network capacity are shared with the other servers on the node. We would rather say that plainly than imply dedicated metal.

Where exactly is the server?

Helsinki, Finland — Hetzner availability zone hel1-dc2, which we read from the instance metadata rather than taking from a marketing page. It is our only location today, so if you are playing from outside Europe, expect the latency that distance implies.

Is the storage really NVMe?

Hetzner publishes NVMe SSD for its Cloud servers, and from inside the instance we can confirm the disk is non-rotational and supports TRIM. What we cannot do is see the physical device through the virtualisation layer, so we are sourcing that claim to Hetzner rather than measuring it ourselves.

What happens to my world if something breaks?

Server data is backed up with Borg to a Hetzner Storage Box that is separate from the servers themselves. The control host backs up every six hours and the game node nightly, and we have tested restoring from those backups rather than assuming they work. You can also take your own backups from the panel at any time.

Do you guarantee uptime?

We do not publish an SLA, because we are not yet in a position to back one honestly. We run one node in one location, so there is no failover to another machine today. Monitoring and alerting are in place, and the refund policy applies, but a promised uptime percentage would be a number we invented.

Why are some plans out of stock?

Because we sell only what we can actually run. The node has a fixed amount of memory available to customer servers, and when that is committed, larger plans stop being orderable until we add capacity. It is deliberately better than taking your money and failing to build the server.

Can I trust your benchmark numbers?

Check them rather than trust them. Every figure we publish names the software version, the world seed, the settings, how many times we repeated the run and what the range was. They are measured on an internal test machine that is not the production node, and every page carrying a number says so beside the number.

Ask us anything this page does not answer

If there is a fact about the infrastructure you want before ordering, ask. If we do not know it, we will say so rather than estimate.

Compare Minecraft plans