A server can advertise plenty of RAM and still stutter the moment five players fly into fresh terrain. That is why Minecraft hosting benchmarks have to measure the work Minecraft actually does: keeping ticks on time, generating chunks, saving the world and running plugins or mods while players are active.
For a private SMP, a public Paper server or a heavy Forge modpack, the right benchmark is not a single number. It is a repeatable test that shows where performance falls apart. CPU behaviour matters more than headline core counts, storage affects saves and chunk loading, and network quality decides whether players can actually feel that performance from where they live.
What a Minecraft hosting benchmark should measure
Minecraft targets 20 ticks per second. A tick is the server's regular update cycle for entities, redstone, world logic and player actions. If every tick finishes within 50 milliseconds, the server holds 20 TPS. Once ticks regularly take longer, TPS drops and players see delayed block breaking, rubber-banding, slow farms and mobs that seem to freeze.
The most useful numbers are TPS and MSPT — milliseconds per tick. TPS tells you whether the server reaches its target. MSPT shows how much headroom remains before lag starts. A server holding 20 TPS at 48 MSPT is technically healthy, but it has almost no reserve for a busy event, new chunk generation or another player joining with a large modded base.
A fair benchmark also tracks memory use, CPU load, disk activity and network latency. These numbers give context. More RAM may prevent crashes caused by a modpack's memory demand, but it will not fix a CPU-bound server spending 80 milliseconds per tick — our guide on fixing low TPS walks through telling those cases apart.
TPS is the result, not the diagnosis
TPS is what players feel, so it deserves attention. But a flat 20 TPS with one person standing at spawn tells you very little. Good testing includes normal behaviour: exploring, teleporting, loading farms, running redstone, fighting mobs and moving between dimensions.
On Paper or Purpur, spark shows exactly which plugins, entities or chunks eat tick time — run /spark profiler during a busy session and read the report. Spark also works on Forge, NeoForge and Fabric, so modded servers can use the same approach. Test the gameplay your community will actually create, not an empty world that makes every host look fast.
Why single-thread CPU speed usually decides performance
Minecraft has become better at spreading some work across threads, especially chunk generation and networking. Still, the core game loop depends heavily on single-thread CPU performance. A processor with many slower cores can lose to a newer CPU with fewer, faster cores once the main server thread is loaded.
This is most visible on Paper, Purpur and Vanilla servers with active players, entity-heavy farms or complex redstone. It matters just as much in modpacks, where machine networks, automation, world generation and mod logic all add tick work. The number to watch is sustained tick performance under load, not a provider's maximum clock or total core count.
Consistency matters as much as peak speed. A node can look great when it is quiet and struggle at peak hours when many demanding servers are busy at once. A single benchmark run cannot prove long-term quality, so repeat it at different times — especially when your players normally log in.
Run /spark tps during your busiest hour: it shows TPS and MSPT for the last few seconds, minute and longer windows. If the worst MSPT values keep creeping towards 50, the server has run out of headroom even while TPS still reads 20.
Storage: why NVMe is more than a spec-sheet line
Storage rarely fixes a bad TPS problem on its own, but slow or overloaded disks cause their own kind of pain. Minecraft constantly reads region files, writes player data, saves chunks, creates backups and processes new terrain. During a save spike, weak storage can delay the main thread enough for players to notice.
NVMe storage shortens the wait for these operations. The biggest gains show up when players explore new areas, when a large world saves and when a modpack writes many files frequently. It also makes restarts and restores far less painful for admins.
Benchmark storage the Minecraft way. Pre-generate a realistic area, restart the server, teleport between distant regions and watch MSPT while the world saves. A raw disk speed result is useful background, but it cannot replace watching the game server under real disk activity. If chunk loading is the bottleneck, see our tips on reducing chunk loading.
Network tests need more than a ping number
Low ping is valuable, but it is not the whole story. Players also need stable routing, low packet loss and enough capacity when several people are online. A server can show a respectable average ping while occasional packet loss causes rubber-banding or disconnects.
Put the server close to the largest share of your players. For a group split between North America and Europe there is no perfect spot: decide which side gets the lowest latency, then check what the other side can expect. Communities with a concentrated audience should place the server near that audience rather than chasing the cheapest price in a distant data centre. Our guide on choosing a server location covers this in detail, and the cities we offer are listed on the locations page.
Test from real player connections whenever you can. Ask a few players in different regions to join during an active session and report movement delay, chunk loading and disconnects. That feedback is worth more than any single latency figure.
How to run a fair benchmark
Start from a clean, known configuration. Record the Minecraft version, server software, Java version, view distance, simulation distance, allocated memory, plugin list and modpack version. If any of these change between tests, the comparison loses meaning. Remember that Minecraft 26.x needs Java 25, so keep the Java version identical across hosts too.
Use the same world seed and test route on every host. Pre-generate part of the world if your real server will, but also include a separate fresh-terrain test. Then simulate your expected load: have players explore, run a typical farm, enter the Nether or End and use the systems your server is built around. A creative-building SMP and a combat-heavy modded community stress the server in very different ways.
During each session, record average and worst MSPT, TPS, memory use, restart time and any visible lag. Run the test long enough to capture autosaves and normal garbage collection. Ten quiet minutes are not a serious result; an hour of ordinary gameplay is far more revealing.
If you are comparing plans, raise the player count or workload gradually. The goal is not to force a crash — you are looking for the point where MSPT consistently approaches 50 and gameplay loses its safety margin.
Reading the results without falling for easy claims
More RAM is not automatically better. Give the server enough memory for its version, modpack, player count and overhead — our guide on RAM for modded servers helps size it — but don't treat RAM as a substitute for CPU. A very large Java heap can even make garbage collection pauses harder to manage on a poorly tuned server.
Likewise, "unlimited players" is not a meaningful promise. The practical player count depends on world settings, server software, player behaviour, mods, plugins and how much terrain each player generates. Twenty players building in one town can be easier to host than six players flying in different directions through unexplored modded terrain.
A good host is open about the workload a plan supports and gives you tools to manage it. A Pterodactyl panel turns routine work — editing files, reading logs, restarting, scheduling tasks — into a few clicks. Automatic backups matter as much as raw speed when an update, a mod conflict or an accidental command damages the world.
At Elysium, Minecraft servers run on high-frequency Ryzen CPUs and NVMe storage with DDoS protection, a nightly automatic backup and 24/7 support. That does not mean every server needs the biggest plan — it means a Vanilla SMP, a plugin community or a modded world gets a stable base to grow on.
Benchmark for your players, not for a marketing chart
The best host for a small friends-only world is not necessarily the best choice for a public survival server with events, custom plugins and players across several time zones. Test the version, modpack and player behaviour you actually plan to run, and keep enough TPS and MSPT headroom for the messy moments: a new update, a huge build, a weekend event or a group heading into fresh chunks together.
When the numbers match how your community plays, hosting stops being something you troubleshoot every night — and your attention goes back to playing, building and keeping the world alive.
Elysium runs Minecraft servers on high-frequency Ryzen and NVMe with DDoS protection and a nightly backup included. The server core is chosen in the order form, and plans and prices are on the Minecraft page.