How much RAM a modded Minecraft server actually needs

Realistic figures

These are heap, not machine memory. The machine needs meaningfully more.

Never give the JVM everything

The single most common mistake: a 8 GB box with -Xmx8G. The JVM needs memory outside the heap — metaspace, thread stacks, direct buffers, the JIT's own code cache — and the operating system needs some to function.

When it runs out, the kernel kills the entire process. You get no crash report, no exception, just a log that stops mid-sentence. See no crash report.

Cap -Xmx at roughly 75% of physical RAM.

Set -Xms equal to -Xmx

Letting the heap grow gradually causes a class of GC stall during world generation. Setting them equal avoids it — and has the useful side effect of failing immediately if the value is impossible, rather than an hour in.

More RAM is not always the fix

This is the part worth internalising. If raising the heap only delays the crash, you do not have a size problem — you have a leak, and more memory buys proportionally more time before the same ending.

How to tell them apart:

• Too small: dies under predictable load — a big base, many chunks loading, several players. Raising it fixes it permanently.
• A leak: survives a restart, then dies again, at a shorter interval each time. Suspect anything holding entity or chunk references — mob mods, chunk loaders, backpacks, anything with a "nearby entities" scan.
• A worldgen spike: dies at the same place every time, with plenty of headroom otherwise. Check what loaded just before.

Watch heap usage across hours, not minutes. A leak is invisible on any shorter scale.

See also java.lang.OutOfMemoryError.

Doing this automatically

Forge Server Doctor performs the whole procedure above — reads the report, finds the real cause under the wrappers, skips the vanilla frames, and names the mod jar. 24 failure modes, each with its fix. The free edition covers the six most common startup failures.