Monolithium

🗿 Finding the Largest Minecraft Alpha Monoliths 🗿

Seed: 26829160 • (x: 0, y: 0) • Area: 1,044,848 blocks squared • First Millionaire
^{Using Moderner Beta Alpha v1.1.2_01 Terrain on MC 1.21 • Distant Horizons • Bliss Shaders}

🔥 Description

Monoliths are a terrain generation bug that happened in the ancient Minecraft Infdev through Alpha 1.1.2_01 versions. They are a rare, often large, flat and tall structures, almost entirely made of stone, with a hidden ocean of water until bedrock beneath them.

Hidden in plain sight for 15 years, I wanted to know how rare they actually are, and find the biggest one!

• This repository contains a brute-force search algorithm in Rust with a basic Python package for statistical analysis, alongside the most interesting findings of the journey.
• Feel free to contribute speed, documentation, cuda port improvements.

Warn: This is a side project, I may have time to port the readme to a mkdocs website in the future.

^{Note: For a more technical explanation, see kahomayo/monolith-renderer 😉}

📦 Installation

Note

Windows users will need to install Visual Studio Build Tools for linkers, or alternatively a MSYS2 environment with the package mingw-w64-ucrt-x86_64-gcc and compile in a UCRT64 shell.

Install Git and astral-sh/uv, open a terminal in some directory and run:

• git clone https://github.com/Tremeschin/Monolithium && cd Monolithium
• Rust part: uv run rustlith (commands)

You can pass any --<feature> explained in Cargo.toml for speedups, like --fast!

🔥 Commands

Warning

• The code will shred your cpu, make sure you have a good cooling solution!
• Large directories are created where you run it - any of target, build, release.

You can run any of rustlith (command) --help for options and information!

🔴 Find all Monoliths in a world

This will search a 8,388,608 blocks square in both positive X and Z directions. Note that all monoliths repeats every such value on any coordinate - there are 9 copies of each within the Far Lands on any given world!

• rustlith search --depth seed --value 617

🟡 Find seeds with spawn monoliths

This will search for seeds that contains monoliths close to spawn.

• Search 0 through 100k seeds: rustlith search linear -t 100000
• Search 50k random seeds: rustlith search random -t 50000

🟢 2Pass heuristic method

This heuristic finds seeds with great potential for large monoliths, by only looking at the much rarer hill noise values, and discarding most seeds with poor x, y, z fractional parts offsets.

• rustlith search --radius 8192 random -t 5000000 --candidates --fast (potential)
• For each output, run rustlith search --step 512 seed -v <n> (real area)

There's a couple improvements to this method:

• Since this discards millions seeds a second, rayon work-stealing parallelism becomes an overhead. Passing --chunks 1000 to search makes each work block process multiple seeds instead of one.
• The hill wraps around 2**19 blocks, so --radius 262144 searches "the whole world"'s potential.
• For long searches, filter out candidates with at least --area n to ignore bad shuffling seeds.
• Largest monoliths are basically guaranteed to hit a lattice point multiple of 1024
• At world::good_perlin_fracts, one can tweak the treshholds for a "good" seed.

Full command idea that broke many records:

# or 'cargo run --release --features candidates --features fast' in monolithium dir
$ QUALITY=250.0 rustlith --candidates --fast \
  search --chunks 1000 --hill --step 1024 --area 2200000 \
  random --total 100000000 \
> candidates.txt

# Many minutes/hours later..
$ rustlith search --depth --step 512 --threaded \
  file -i candidates.txt \
> final.txt

With enough --total seeds, checking the best ones almost guarantees a record :)

🟠 Fast spawn search

Instead of initializing a world from zero per work chunks loop, this method reutilizes most perlin noises data in a incremental way, rolling down all octaves and appending a new one with the current RNG state.

As such, --chunks plays more than one role here, theoretical max performance at total_seeds/cpu_threads:

$ AREA_STEP=8 rustlith --fast --sister-perlin --deque-octaves \
  search --chunks 1000 --area 500000 \
  random -t 50000000

⭐️ Showcase

Note

Area calculations are within 1% error, nearby monoliths are part of the same complex.

🔵 Largest monoliths anywhere

Area	Seed	X	Z	Date	Found by
3,119,151	94116384388573	3185395	1401244	2025/10/06	Tremeschin
3,099,536	1184681998931	7361728	4530048	2025/10/18	Ferra
3,047,088	24388692462	2942560	4232256	2025/10/18	Ferra
3,012,432	1102581481210	6025536	4676000	2025/10/18	Ferra
2,987,360	34786730572	4236288	5081952	2025/10/18	Ferra
2,950,096	530268432158	1384896	8037440	2025/10/18	Ferra
2,892,928	873667255313	5711040	1334432	2025/10/18	Ferra
2,890,848	143779371652733	265920	5994240	2025/09/20	akatz-ai
2,649,984	130449915832690	786608	4163520	2025/09/20	akatz-ai
2,316,064	19907909658842	1572070	3668360	2025/09/18	Tremeschin
2,033,040	250673273362854	7337600	5768320	2025/09/18	Tremeschin
1,992,096	281351900698438	4716910	4718668	2025/09/18	Tremeschin

🔵 Largest monoliths near spawn

Area	Seed	Date	Found by
2,420,304	213457002840989	2025/10/29	NathanVH8
2,240,624	38078992166078	2025/10/19	NathanVH8
1,993,040	269943915147763	2025/10/18	NathanVH8
1,975,712	35668441367715	2025/10/08	NathanVH8
1,745,664	4609608251	2025/08/13	Tremeschin
1,584,112	1847066092	2025/08/13	Tremeschin
1,420,816	2045872561	2025/08/13	Tremeschin
1,371,824	3847304212	2025/08/13	Tremeschin
1,369,360	1593912439	2025/08/13	Tremeschin
1,345,520	4563197188	2025/08/13	Tremeschin
1,305,472	4432659853	2025/08/13	Tremeschin

🔵 Lowest seed visible from spawn

Drum rolls.. 617. This seed contains a visible Monolith from spawn 🤯

🔵 Monoliths repeat every 8,388,608 blocks

For every monolith in a world there's 9x exact copies of them within the Far Lands:

• A monolith at spawn appears on:
• (-x, x) • ( 0, x) • ( x, x)
• (-x, 0) • ( 0, 0) • ( x, 0)
• (-x, -x) • ( 0, -x) • ( x, -x)

Sadly, the Far Lands override the monoliths, there's no such thing as a Far Monolith 😿

Note

For the keen among you, that value is 2**23 - this happens for a couple of reasons:

• Ken Perlin's noise, unscaled, repeats every 256 units on any coordinate • (2**8)
• There are 16 octaves on the hill noise, each octave halves the previous frequency, so the highest one repeats every (2**15) blocks (starting from multiplier 1).
• Minecraft samples every 4 blocks, the depth scale is 100 but & 0xFF truncations cancels it at 2**20 Multiplying factors, 2**(8+15) = 2**23

🔵 How rare are they?

Spoiler: Not much, certain seeds are more likely to generate monoliths (anthropic principle confirmed?), but most contains at least half a million monolith complexes within the Far Lands (12,550,824 blocks squared).

🚀 Speeds

Monolithium is written in heavily parallelized Rust with the help of crates like Rayon for fearless concurrency and Ahash fast hashing, fully utilizing all available CPU resources one throws at it.

🦀 For a Ryzen 9 5900X 12c/24t, 2x3200 MT/s DDR4 CL16 2Rx8 system, one might expect:

• 3.75 minutes to find all monoliths in a seed, probing every 128 blocks.
• Search 1,150,000 seeds per second for spawn monoliths (approximated)
• Search 440,000 seeds per second for spawn monoliths (accurate)

Such speeds scales about linearly with your hardware - for better or worse!

🔎 Future work

• Investigate the correlation of Perlin coefficients to the likeliness and size of Monoliths.
• Is it more efficient for CUDA to stream perlin coefficients than inline JavaRNG on CPU?
• Make statistical analysis (Average size, Distribution) of Monoliths in seeds.
• Make a HashMap<(int, int), Monolith> to avoid recomputing areas
• Throw 2x Epyc 9965 at the code. I have a spare one for sure iirc.

♻️ Credits

• User @kahomayo for monolith-renderer to understand the underlying mathematics.
• YouTuber @AntVenom For the Breaking Minecraft playlist inspiration.