Game Hosting Guides · Independent reviews

How Much RAM Does a Minecraft Server Need? (2026 Guide)

How much RAM does a Minecraft server really need? Honest 2026 numbers for vanilla, plugin, modded, and proxy servers, with concrete buying advice.

By · Last updated · 8 min read

The short answer: a vanilla Minecraft server for five friends needs 2 GB. A Paper server with a dozen plugins and 15 players wants 4 GB. A 100-mod CurseForge pack with ten players wants 8 GB. A 200+ mod kitchen-sink pack with 20 players wants 10-12 GB, and you should worry more about the CPU it lands on than the RAM number on the invoice. Anything sold as needing 16 GB for a tiny vanilla SMP is upsell. Anything sold as 1 GB for a heavy modpack is bait. This guide explains where those numbers come from and how to tell when you genuinely need to upgrade.

The quick lookup table

Server type Player count Recommended RAM Why
Vanilla 1-5 2 GB Server.jar idles around 600 MB; 2 GB leaves headroom for chunk loading and a couple of villager farms.
Vanilla 6-15 3-4 GB More loaded chunks per player, more mobs, more entities.
Vanilla 15-30 4-6 GB Chunk-loading overhead scales roughly with concurrent players.
Paper / Spigot / Purpur 1-10 3-4 GB Plugins add resident memory and async tasks; Paper’s optimizations help but don’t shrink heap.
Paper / Spigot / Purpur 10-25 5-6 GB EssentialsX, LuckPerms, WorldGuard, dynmap and similar staples each carry their own cost.
Paper / Spigot / Purpur 25-50 6-10 GB Plugin servers at this size start to look like small modpacks in memory profile.
Forge / Fabric / NeoForge 1-5, small pack (~50 mods) 4-6 GB Modded JVMs need bigger heap headroom to avoid GC stutter.
Forge / Fabric / NeoForge 5-10, mid pack (100-150 mods) 6-8 GB Pack with Create, Mekanism, Thermal, etc. easily fills 6 GB at startup.
Forge / Fabric / NeoForge 10-20, large pack (200+ mods) 10-12 GB All The Mods, RLCraft, Better MC, kitchen-sink packs.
Forge / Fabric / NeoForge 20+, large pack with chunkloaders 12-16 GB Pre-generated chunks, quarries, large industrial bases.
Proxy (Velocity, BungeeCord, Waterfall) Any 512 MB - 1 GB Proxy just routes packets; backend servers do the work.

These are the numbers we’d actually buy. Hosts will happily sell you more.

Why RAM isn’t the only thing that matters

The most common buying mistake is treating RAM as a proxy for “server power.” It isn’t. Minecraft’s main server tick runs on a single thread, and TPS (ticks per second) is gated by the clock speed of whatever CPU core that thread lands on. You can throw 16 GB at a server running on a 2.4 GHz cloud core and still see 12 TPS on a 20-player Paper world.

This is verifiable in the hardware our comparison page breaks down. Shockbyte’s newer nodes run AMD Ryzen 9 7950X (5.7 GHz boost) and EPYC 4465P / 4244P, but older nodes still use Intel Xeon E-2276G — and which one you land on isn’t disclosed at checkout. Hostinger’s Game Panel VPS uses AMD EPYC Milan (e.g. EPYC 7543P, 2.8 GHz base / 3.7 GHz boost) — good silicon, but per-core clocks are middling next to a desktop Ryzen. Server.pro is unusually honest, splitting its fleet into three labeled tiers — EPYC 7351P on the free and entry “Hosting” plans (a weak 2.4 GHz base), Ryzen 7 5800X on “Gaming”, and Ryzen 9 9950X3D with 3D V-Cache on “Performance”.

For Minecraft, single-thread clock and L3 cache matter more than core count. A 4 GB plan on a Ryzen 9 7950X will outperform an 8 GB plan on a contended cloud vCPU every time. If a host won’t tell you what CPU you’re getting, that’s a buying signal. Buy RAM for memory pressure. Buy CPU disclosure for TPS.

RAM for vanilla servers

Vanilla is the cheapest case and the one most often oversold. The Java server JAR for Minecraft 1.21.x starts around 500-700 MB resident once spawn chunks load. Each additional concurrent player adds roughly 50-150 MB depending on render distance, server view-distance, and how spread out players are. Entities are the other cost — mob farms, large animal pens, item-frame walls all carry persistent memory cost.

Sweet spot: 2 GB for under 5 players, 3-4 GB for 5-15 players. Going from 2 GB to 4 GB on a 5-player vanilla server will not raise your TPS. It gives you more headroom before the garbage collector has to do a big pause — on small servers, rare enough you won’t notice.

For context, Shockbyte’s 1 GB Dirt plan at $2.50/mo is technically a working vanilla server (documented as 3-5 players) but gives you almost no headroom for a single overzealous spawner. The 2 GB Sand plan at $5.00/mo is the more honest entry point for a real friend group.

RAM for plugin servers (Paper, Spigot, Purpur)

Paper, Spigot, Purpur and their derivatives are heavily optimized vanilla forks. They reduce tick cost, improve chunk loading, and asynchronously handle things vanilla blocks on. But plugins still have to live somewhere, and that somewhere is the JVM heap.

Rule of thumb: a plugin server wants roughly 2x the RAM of a vanilla server at the same player count. That sounds dramatic — it isn’t. Five typical plugins (EssentialsX, LuckPerms, WorldEdit, WorldGuard, Vault) idle around 300-500 MB combined once their data stores are warm. Heavier plugins (dynmap, BlueMap, CoreProtect, plot management, economies with large databases) can each add 200-500 MB.

A 10-player Paper SMP with the EssentialsX starter set lives comfortably in 4 GB. Adding dynmap, CoreProtect with months of history, and a couple of minigame plugins pushes you to 6 GB. Beyond that on a small plugin server you’re not buying performance, you’re buying GC headroom — useful, diminishing returns.

Paper does a lot of work to keep tick cost down on cheap hardware. If you’re cost-constrained, Paper on a 4 GB plan with a good CPU will outrun vanilla on an 8 GB plan with a slow CPU, every time.

RAM for modded servers (Forge, Fabric, NeoForge)

This is where buying decisions get consequential, and where most undersizing failures happen. Modded Minecraft has a different memory profile:

A modest 50-mod Fabric pack with 4-5 players runs cleanly in 4 GB. A 100-mod CurseForge pack like a current All The Mods entry idles around 4-5 GB before anyone connects and wants 6-8 GB for 5-10 players. Larger packs — RLCraft, Better MC, the bigger ATM / FTB packs at 250-400 mods — routinely need 8-12 GB and will OOM-crash on anything smaller during worldgen or a mid-game base teleport.

Garbage collection is the second variable. Java’s default GC is fine for small heaps; on a modded server with a 10 GB heap and the short-lived allocations modded Minecraft generates constantly, defaults produce long pauses that show up in-game as freezes. The community standard is Aikar’s flags — a tuned G1GC configuration most decent hosts apply for you, but worth verifying in your startup command if you see pauses on a properly sized server.

For modded specifically, our review of BisectHosting Premium holds up: higher-clock CPUs and lower node contention matter as much as the RAM number. Buying a 12 GB Budget plan to “save money” on a modpack server is false economy — you’ll spend the saving on Discord arguments about lag.

RAM for proxy/network servers (BungeeCord, Velocity)

If you’re running a network — a lobby, survival world, creative world, and minigames world behind a single connection address — the proxy itself needs almost nothing. BungeeCord, Waterfall, and Velocity are packet routers; they don’t load chunks, run physics, or tick entities.

Allocate 512 MB - 1 GB to the proxy. Even large networks routing thousands of concurrent connections rarely need more than 2 GB on the proxy itself. The RAM that matters is on each backend server.

The proxy should be a separate instance from your backends, not co-located. Most managed hosts (Apex Hosting exposes Velocity/BungeeCord/Waterfall as first-class options in the panel) let you spin up a small dedicated proxy plan cheaply. If you’re already paying for an 8 GB backend, don’t squeeze a proxy onto the same JVM — you lose the isolation that’s the point of running a proxy.

Common RAM-buying mistakes

A few patterns that come up over and over in our review research:

  1. Overpaying for 8 GB on a shared, slow CPU. The most common failure mode. More RAM does not fix CPU contention. If your TPS is bad and memory usage is at 40%, more RAM will not help.
  2. Undersizing modpack servers. Hosts will let you buy 4 GB and install a 200-mod pack. It’ll start, then crash on world load or first big base render. Check the pack’s documentation for recommended server RAM and treat it as a floor, not a target.
  3. Allocating all of the host’s RAM to the JVM. The OS needs RAM too — kernel, network buffers, the panel agent, scheduled tasks, sometimes a backup process. On a 4 GB plan, -Xmx4G is asking for an OOM kill. Managed hosts cap this for you; on a VPS like Hostinger’s Game Panel, you leave ~512 MB to the OS yourself.
  4. Confusing “free tier RAM” with usable RAM. Server.pro’s free tier offers ~1 GB on shared CPU with hourly renewal — a testing environment, not production, and they’re honest about it. Aternos and similar pure-free hosts sleep when nobody’s online. Free RAM is rarely 24/7 RAM.
  5. Buying for peak load you don’t have. A 20-slot server averaging 4 concurrent players doesn’t need to be sized for 20. Buy for your realistic average, and use a host that upgrades in a few clicks.

How to actually measure if you need more RAM

Don’t guess from forum advice — measure your own server. Three ways:

The honest version: if you play comfortably and /tps shows ≥19.5 sustained, you have enough RAM. Don’t upgrade prophylactically.

RAM recommendations by hosting plan

Mapping the numbers above to real plans we’ve reviewed:

For side-by-side pricing across providers at every RAM tier, our hosting comparison page breaks down what each host charges per gigabyte and which ones disclose their CPUs. RAM is the easy number on the invoice; the rest of the spec sheet is what determines whether the server is actually good.