How much RAM does a game server need? 2026 cross-game sizing guide

If you've shopped for game server hosting in the last year, you've seen the same pattern: providers list RAM tiers from "2 GB / €3" to "32 GB / €60" with vague hints about "supports up to 50 players". The numbers are honest in isolation, but they don't tell you which tier you actually need. So you over-buy.

This post is the cross-game version of the conversation we keep having in support: "I want to host X for my Y-person group, what should I pick?" We host 17 different games at Renzom, so the numbers below come from real container metrics. No guessing, no upsell — same advice we'd give a friend who asked.

The 30-second answer

Pick the game, pick your group size, pick the RAM number.

Game	Small (2–4)	Medium (5–10)	Large (16–32)	Why it differs
Minecraft (Paper / vanilla)	2 GB	4 GB	8–16 GB	Linear with players + mods
Minecraft (Modded, heavy modpack)	6 GB	8–10 GB	12–24 GB	JVM heap + chunk caching
Hytale	4 GB	6–8 GB	10–16 GB	Java 25, persistent biome state
Palworld	8 GB	12 GB	16–24 GB	Pal AI is RAM-heavy
Valheim	4 GB	6 GB	8 GB	Capped at 10 players; very efficient
Enshrouded	6 GB	8 GB	12–16 GB	16-player cap, voxel world
ARK: Survival Ascended	12 GB	16 GB	24–32 GB	UE5, brutally RAM-hungry
ARK: Survival Evolved	8 GB	12 GB	16–24 GB	UE4, lighter than ASA
Rust	8 GB	12 GB	16–24 GB	Wipe-day spikes drive sizing
7 Days to Die	6 GB	8 GB	12–16 GB	Voxel + horde nights
Project Zomboid	4 GB	6 GB	8–12 GB	Java heap + save bloat
Satisfactory	8 GB	12 GB	16+ GB	Late-game megafactories balloon
Factorio	4 GB	4 GB	8 GB	Single-thread, UPS-bound not RAM
Terraria	2 GB	2 GB	4 GB	Tiny footprint, very efficient
CS2 (5v5 competitive)	2 GB	4 GB	4 GB	Tick rate + plugins
Garry's Mod	2 GB	4 GB	6 GB	Source engine, addon-driven
Subnautica 2	4 GB	6 GB	8 GB	8-player cap, oceanic streaming
Windrose	8 GB	12 GB	16–24 GB	Ship physics + world tiles
Discord bot	256 MB	512 MB – 1 GB	2 GB+ per shard	Bot type defines this, not slot count

That's the answer for most people. The rest of this post is the why — useful if you want to scale carefully, run something modded, or argue with your friend who insists you need a 32 GB box for 4 players.

The three categories that actually drive RAM

Once you host enough games, you stop thinking about them as 17 separate engines and start seeing three families.

Family 1: Linear with slots

The pure case is Valheim, Terraria, CS2. RAM use scales roughly proportionally with concurrent players. Double the slots, roughly double the active state. These games are the easiest to size: count your players, pick the tier, done.

Why are they so well-behaved? Their world state is either small (CS2 — one map, no persistent objects) or efficiently bounded (Valheim — fixed seed, biomes load on demand, world is small by AAA standards). The JVM or game-engine cost stays predictable as you add players.

Family 2: Quadratic with player spread

The painful case is Hytale, Minecraft (modded), Windrose, Enshrouded, Subnautica 2. RAM doesn't just scale with players — it scales with how spread out the players are.

Two players standing on the same chunk: 1 active region, 1× the RAM. Two players on opposite ends of a 4 km world: 2 active regions, ~2× the RAM. Eight players exploring eight different biomes: 8 active regions, ~8× the RAM — even though it's only 4× the player count.

This is why a 4-player Hytale server can need 4 GB (clustered) or 8 GB (scattered) depending on play style. Survival sandbox games where exploration is the loop have to keep simulating any region a player is in, plus a buffer around it for entity AI and physics.

Heuristic: pick the tier for "all players actively exploring" rather than "all players at base". You can always shrink later.

Family 3: Compute-bound, not RAM-bound

Factorio is the canonical case. The bottleneck is UPS (updates per second) on a single CPU thread, not RAM. A 4-player Factorio megabase can sit at 4 GB resident forever, but TPS will degrade because one of your CPU cores is pegged at 100%.

CS2 is similar — 128-tick competitive servers are CPU-bound on the tick loop. RAM almost doesn't matter past 2 GB. If you're seeing rubber-banding, you don't need more RAM, you need a single-thread-fast CPU.

Same for Rust on wipe day: the bottleneck is single-thread tick performance under high entity load, not RAM. Adding RAM to a struggling Rust server doesn't help; moving to a host with stronger single-thread CPU does.

How to tell which family you're in: open your panel's resource graph. If CPU saturates before RAM, you're CPU-bound — buy CPU, not RAM. If RAM saturates first, you're in family 1 or 2.

Modded games deserve their own line

Every column in the table above assumes vanilla or light modding. The moment you load a 200-mod modpack:

• Minecraft (modded): Add 4 GB to the vanilla tier for a 50-mod pack. Add 8 GB for a 200-mod pack. ATM 9 with 12 players on 4 GB will OOM in 20 minutes.
• ARK with mods: Mods can double or triple RAM use. A 16 GB ASA tier can struggle with 5 mods if they introduce custom dinos with high entity counts.
• Hytale with plugins: Each plugin keeps state in the JVM heap. 10 plugins ≈ +1 GB. 30 plugins ≈ +3 GB on top of vanilla sizing.
• Project Zomboid with Workshop mods: Java heap inflates fast; the save folder itself can grow to several GB. Plan for 8 GB minimum if you run Workshop content.

The general rule: if you're modding, take the recommended vanilla tier and add 50%. Then watch your panel graphs for a week and adjust.

Common mistakes that waste money

Buying the biggest tier "just to be safe". A 32 GB Minecraft tier for 4 friends is 4× what they need. They never use the headroom, the game runs the same as on 8 GB, and you pay €40–60/month extra for nothing.

Sizing for the announcement, not the reality. The publisher's marketing said "supports 64-player worlds!" Your friend group is 5 people. Size for the reality.

Mixing CPU-bound and RAM-bound thinking. Factorio doesn't need more RAM, it needs a faster single-thread CPU. CS2 same. If your panel shows 90% CPU on one core and 30% RAM use, more RAM doesn't help.

Ignoring the slot-vs-spread split. A 16-slot Hytale server with 4 active players is not the same load as a 16-slot Hytale server with 16 active players spread across the world. Size for the realistic peak, not the marketed cap.

Refusing to upgrade once you're full. RAM upgrades on most managed panels are live — change the tier, restart, done. Don't suffer through TPS drift or OOM kills because "we can manage". The €5/month difference between 8 GB and 12 GB tiers is cheaper than your friends quitting because the server keeps crashing.

When to scale up

Watch your panel resource graph for these signals:

• RAM: Sustained >85% utilization across an hour. Upgrade one tier.
• CPU: Sustained >85% on the busiest core (look for single-thread saturation specifically). Upgrade CPU, not RAM.
• TPS / tick drift: Game-specific, but most panels expose it. Sustained drift > 10 ms is your warning shot.
• OOM kill in panel logs: Already too late. The container died. Upgrade before this happens.
• Restart-time validation slows down: Disk pressure. Either upgrade tier (which sometimes moves you to a faster NVMe) or change hosts.

Per-game deep dives

For more on specific games we host, see:

• Hytale RAM sizing 2026 — Java 25 specifics, plugin sizing
• Discord bot RAM sizing 2026 — by bot type
• Palworld dedicated server settings deep-dive 2026 — settings that affect RAM
• Windrose dedicated server setup 2026 — SteamCMD + NAT punch-through
• Best new game servers 2026 — what's worth hosting right now

If you want to skip the research and just configure with a live tier estimate, our configurator takes RAM, slot count and game and gives you the price in real time. Same numbers as this post, just clickable.

That's the honest cross-game version. The marketing says "more is better"; the metrics say "right-sized is better".