# How much RAM do Minecraft modpacks really occupy?

> A measurement series with 37 modpacks on the same test server

> NOT AN OFFICIAL MINECRAFT PRODUCT. NOT APPROVED BY OR ASSOCIATED WITH MOJANG OR MICROSOFT.

- Publisher: Nexthosting
- Authorship: Erik Friedl (ORCID: https://orcid.org/0009-0008-2784-4314), Nexthosting
- DOI: https://doi.org/10.5281/zenodo.24430340
- Type: Own measurement
- Category: Game servers
- Version: 1.0.0, As of Nov 21, 2026
- Period: Nov 21, 2026
- Sample: 35 of 37 modpacks evaluated
- License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/deed.en)
- Original page: https://nexthosting.net/en/studies/minecraft-modpack-ram
- PDF: https://nexthosting.net/en/studies/minecraft-modpack-ram/studie.pdf
- Translation of the German original. The original is authoritative. https://nexthosting.net/studien/minecraft-modpack-ram

## Summary

Our own measurement: memory of Minecraft modpacks after startup and with a loaded world, with method, raw data and checksums.

## Research question

How much memory do widely used Minecraft modpacks actually occupy on a server, and how well does the number of mods predict this value?

## Key results

- **Persistently occupied memory (median)** (0.77 GB): Half of the 35 fully measured packs occupy less than 0.77 GB with 256 loaded chunks, the other half more. Measured after garbage collection and without players. This is not the server size you need: Java needs additional space, and so do players.
- **Peak of the whole process (median)** (2.56 GB): The peak of the Java process had a median of 2.56 GB. It also includes memory for code, threads, buffers and heap not yet cleaned up. The value depends on the Java startup settings (1 to 8 GB heap) and is not a recommended RAM size.
- **Highest persistently occupied memory** (1.78 GB): Creat : More Attachments and Optimization (88 mods) occupied the most memory among the fully measured packs.
- **Lowest persistently occupied memory** (0.30 GB): RealisticCraft (33 mods) occupied the least memory.
- **Relationship between number of mods and memory** (0.65): In this measurement, packs with more mods mostly occupied more memory, with some exceptions. For example, Creat : More Attachments and Optimization with 88 mods occupied 1.78 GB, Elysium Days with 297 mods only 0.86 GB. The value is the Spearman rank correlation across 35 packs (1 would be a perfect match). With so few packs it is uncertain. A relationship is not a cause.

37 modpacks were selected. 35 of them are fully evaluated: each ran on its own on the same test server and was measured twice, after startup and with 256 loaded chunks. The other 2 packs: 1 could not be measured in our setup and 1 runs on a Minecraft version without /forceload, and its world value is only comparable to a limited extent. 7 of the evaluated packs ran only with a documented intervention, for example without client-only mods.

What we measured is the used memory, that is, the heap after two forced garbage collections plus Metaspace. This is the part a pack occupies persistently, and not a recommended RAM size. Player load is not included, so real servers need more.

## Method

- **Install**: The pack is loaded in a fixed version, and every file is checked by SHA-512. Client-only mods are removed. Libraries that a server mod requires according to its mod file stay in. A bundled “Missing Mods Checker” and the mod “CIT Resewn” are removed because they abort the server start in our setup.
- **Start**: The server starts in its own container without network and without players. After the “Done” message we wait 60 seconds.
- **First measurement**: We force two garbage collections (jcmd GC.run) and read heap and Metaspace with jstat. This is the “after startup” value.
- **Load the world**: With /forceload we load 256 chunks (16 × 16) around the origin. This is finished when all 256 are verifiably in the region files, after 5 minutes at the latest.
- **Second measurement**: After a 30-second wait, two garbage collections again, then heap and Metaspace. This is the “with loaded world” value.

### Measurement setup

- Loaded chunks: 256 (16 × 16)
- World seed: 20261121
- Players: none
- Processor: Intel Core i5-13500
- CPU cores: 4 efficiency cores, up to 3.5 GHz
- Maximum heap: 8 GB (start: 1 GB)
- Memory limit of the container: 10 GB
- Garbage collection: G1, 2 × before each measurement
- Java: 8, 17, 21, 25, depending on the pack
- View distance / simulation distance: 10 / 8 chunks
- Network during the measurement: none
- Abort at startup: after 20 minutes
- Unit: 1 GB = 1,024 MB (calculated in binary)

## Sample

The packs on our modpack pages, that is, widely used packs that can be set up on a server. Not a random selection. 1 could not be measured in our setup, 1 runs on a Minecraft version without /forceload, and its world value is only comparable to a limited extent. A start that fails in our setup says nothing about the quality of a pack.

## Limits of what the study can show

- **No players**: Players load new chunks, move objects and fill inventories. We did not measure this. Real servers need more than what is shown here.
- **Small, young world**: We measured a freshly generated world with a fixed seed and 256 chunks. Built machines, farms and worlds played for a long time need more.
- **One run per pack**: Each pack was measured once. Startup times also depend on the load on the measurement machine, where other servers were running alongside, and are only rough comparison values.
- **Fluctuating individual values**: For 3 of 35 packs, the value with the loaded world was below the value after startup. We did not investigate the cause. Individual values should therefore be read with caution.
- **Fixed setup**: All runs had 4 efficiency cores of an Intel Core i5-13500 (up to 3.5 GHz) and at most 8 GB heap. On faster cores, startup times may be shorter. With other Java settings, memory can look different.
- **Used memory is not total consumption**: Java occupies additional memory for code, threads and buffers. The peak of the process is in the raw data, but it depends on the Java startup settings and is not suitable as a recommended RAM size.
- **Packs outside the key figures**: Better MC [FORGE] - BMC4 (about 2.3 million downloads, rank 4 on the list) could not be measured. The required mods twilightforest, connectormod and flib are missing from the Modrinth package. Adding them would change the content of the pack. What was tried: the missing required library TemporalAPI 1.6.6 was added from Modrinth (Extra Compat requires 1.6.5 up to below 1.7.0). Winter Zombies runs on Minecraft 1.12.2, which does not know /forceload. The 256 target chunks were generated with /forge generate and counted afterwards, but do not stay permanently loaded. The world value is therefore only comparable to a limited extent. These packs are not included in the key figures.
- **Interventions on individual packs**: the horrors in the fog: client mod Replay Mod removed (Modrinth: server_side=unsupported). Homestead: watchdog switched off (max-tick-time=-1), because otherwise the server is terminated while loading the 256 chunks. Zombie Storm 100 Days: client mod Just Enough Characters removed (loads client classes, crash on the server). Client mod Fog Overrides removed (loads client classes, crash on the server). Isle of Berk (Claws of Berk): client mod ToroHealth removed (Modrinth: server_side=unsupported). Client mod Fastquit removed (Modrinth: server_side=unsupported). Client mod Enhanced Block Entities removed (Modrinth: server_side=unsupported). FOMFate's Machinery: client mod ReForgedPlay removed (brings the client graphics library lwjgl-tinyexr, which aborts the server start). HORIZONS: Distant Horizons 2.2.1 removed (according to the changelog, support for dedicated servers only exists from 2.3.0). AnimateCraft: Lankeren: library CompleteConfig 2.5.0 replaced by 2.5.2 (2.5.0 aborts on the server with NoSuchFieldError).
- **No random selection**: We measured widely used packs from our range. The results do not apply to every modpack.
- **Fixed versions**: New versions of a pack can differ. The measured version is in the raw data.

## Data and evidence

- [daten.csv](https://nexthosting.net/studien/minecraft-modpack-ram/dateien/daten.csv): Measured values per pack (CSV) (SHA-256 `d9682ed3555b7b7629a00d524916d8e482f973b41bdde8bc87b42d43adb9ff8f`)
- [daten.json](https://nexthosting.net/studien/minecraft-modpack-ram/dateien/daten.json): Measured values per pack (JSON) (SHA-256 `956a72f2d8057d6884a72bc53ebd74d049ad12af6eb577212ab059dafd071e47`)
- [messung.py](https://nexthosting.net/studien/minecraft-modpack-ram/dateien/messung.py): Measurement script (SHA-256 `a223ddde7fbf1cf2715a30cf313218335c12c347a57a5d4388ce05c82325ba97`)
- [packliste.json](https://nexthosting.net/studien/minecraft-modpack-ram/dateien/packliste.json): List of packs (SHA-256 `93319aedf7deb1e9cd49b220b306a40142ff2feaa990645e0ce30e5cb248023f`)

Checksums: https://nexthosting.net/studien/minecraft-modpack-ram/pruefsummen.txt

With the measurement script and the pack list, the measurement can be repeated on a Linux server with Docker. Packs are loaded in the same versions and checked by checksum. Different hardware, Java builds and the load on the machine can change startup times considerably and memory slightly.

### Fields of the data files

- `slug`: Identifier of the pack
- `name`: Name of the pack
- `version`: Measured version of the pack
- `minecraft`: Minecraft version
- `loader`: Mod loader
- `loader_version`: Version of the mod loader
- `java`: Java major version
- `mods` (Anzahl): Mod files on the server after removing client-only mods
- `status`: gemessen, beim_laden_beendet (started, was terminated while loading the world), nicht_gestartet (did not start in our setup), abgebrochen (measurement machine busy) or offen
- `ausgewertet`: true = both measured values exist and all 256 target chunks were permanently loaded. Only these rows go into the key figures.
- `start_gemessen`: true = there is a measured value after startup
- `eingriffe`: Changes to exactly this pack so that it runs in our setup. Empty = none.
- `grund`: Why a pack could not be measured
- `welt_ergebnis`: loaded (all 256 chunks), loaded, with intervention, via /forge generate (generated and counted afterwards, not permanently loaded, only comparable to a limited extent), ended while loading or not finished.
- `start_s` (s): Time from container start to the “Done” message
- `boot_heap_mb` (MB): Heap after two forced garbage collections, after startup
- `boot_metaspace_mb` (MB): Metaspace after startup
- `boot_mb` (MB): boot_heap_mb plus boot_metaspace_mb
- `world_heap_mb` (MB): Heap after two forced garbage collections, with loaded world
- `world_metaspace_mb` (MB): Metaspace with loaded world
- `world_mb` (MB): world_heap_mb plus world_metaspace_mb. This is the used memory, not a recommended RAM size. Empty = not measured.
- `peak_anon_mb` (MB): Highest memory of the process during the run. Depends on the Java startup settings and is not a recommended RAM size.
- `chunks_s` (s): Time until all 256 target chunks were in the region files or the time limit was reached
- `chunks_gezaehlt` (Anzahl): Target chunks that are verifiably in the region files (target: 256)
- `chunks_stabil`: true = all 256 target chunks are in the region files. false = not all target chunks loaded, so there is no world value.
- `forceload_bestaetigt`: The server confirmed the loading of the chunks in the log. false does not automatically mean “not loaded”, some packs log only a raw language key.
- `weltdatei_mb` (MB): Size of the region files of the world
- `bibliotheken_behalten` (Anzahl): Files marked as client mods that had to stay because a server mod requires them
- `gemessen_am`: Day of the measurement

## Sources

1. Modrinth: Modrinth API Documentation. https://docs.modrinth.com/api/ (Retrieved on Nov 21, 2026)
2. Oracle: The jstat Command (Java SE 21). https://docs.oracle.com/en/java/javase/21/docs/specs/man/jstat.html (Retrieved on Nov 21, 2026)
3. Oracle: The jcmd Command (Java SE 21). https://docs.oracle.com/en/java/javase/21/docs/specs/man/jcmd.html (Retrieved on Nov 21, 2026)
4. Minecraft Wiki: Commands/forceload. https://minecraft.wiki/w/Commands/forceload (Retrieved on Nov 21, 2026)
5. Pterodactyl: Yolks: Container-Abbilder für Java. https://github.com/pterodactyl/yolks (Retrieved on Nov 21, 2026)

## Own interest

Nexthosting sells Minecraft and modpack servers itself and has an interest in this study being read and linked. The measurements were made on a Nexthosting server. The raw data and the measurement script are open, so the results can be checked independently of us.

## Changes and corrections

- 1.0.0 (Nov 21, 2026, Version): First complete version.

## How to cite this study

Friedl, E. & Nexthosting. (2026). How much RAM do Minecraft modpacks really occupy? (Version 1.0.0) [Report]. Nexthosting. https://doi.org/10.5281/zenodo.24430340

```bibtex
@techreport{nexthosting2026minecraftmodpackram,
  author     = {Erik Friedl and {Nexthosting}},
  title      = {How much RAM do Minecraft modpacks really occupy?},
  institution= {Nexthosting},
  type       = {Study},
  year       = {2026},
  version    = {1.0.0},
  url        = {https://nexthosting.net/en/studies/minecraft-modpack-ram},
  urldate    = {2026-11-22},
  language   = {english},
  doi        = {10.5281/zenodo.24430340}
}
```
