Benchmarks — MC 1.21.1 w/ Fabric

(Benchmarks ran on the same seed)

Atlas

A Fabric mod that rebuilds Minecraft's chunk generation pipeline from the ground up.

---

So what is Atlas?

Atlas is a Fabric mod that radically accelerates world generation by replacing Minecraft's single-threaded chunk pipeline with a parallel DAG (directed acyclic graph) scheduler, adaptive I/O, and a PID-based throttle controller that keeps your server from falling behind.

Where other mods speed up individual parts of worldgen, Atlas redesigns how chunks are scheduled, generated, and written — treating the entire pipeline as a dependency graph and saturating your CPU cores intelligently instead of just running more threads blindly.

---

Benchmarks — MC 1.21.1 w/ Fabric

Tested on a Dell Latitude E7470 · Intel Core i7-6600U (2c/4t) · 16 GB RAM
Chunky pre-generation · Overworld · Fresh world

Configuration	Avg CPS	Time	Stalls
Atlas	11.33	23m 43s	0.1%
Atlas + Lithium + Spark + Terralith	5.69	12m 23s	4.8%
C2ME 0.3.0+alpha	4.84	55m 31s	11.8%
C2ME + Lithium + Spark + Terralith	3.22	21m 50s	39.7%

Atlas is 2.34× faster than C2ME standalone. Atlas running a full mod stack with Terralith still outperforms C2ME running completely alone — meaning even the worst case for Atlas beats the best case for C2ME on this hardware.

---

How does it work?

⚙️ Parallel DAG Scheduler

Vanilla Minecraft processes chunks sequentially through a fixed pipeline of 12 generation stages. Atlas models these stages as a dependency graph and dispatches independent chunks in parallel, keeping all available CPU cores busy without violating generation dependencies.

🔁 Adaptive PID Throttle Controller

Atlas includes a real-time PID controller (tuned live during generation) that monitors server tick health and automatically adjusts generation throughput. If the server starts lagging, Atlas backs off. When headroom is available, it pushes harder. Your TPS stays smooth even during heavy pre-generation.

💾 Async I/O Pipeline

Region file writes are batched and processed on a dedicated I/O thread, decoupling disk writes from the generation thread pool. This prevents I/O bottlenecks from stalling worldgen, especially on slower storage.

🧵 NUMA-Aware Threading

Atlas detects your CPU topology and assigns work to threads that share cache with the data they're processing — reducing cache misses and memory bus contention on multi-socket or hybrid-core hardware.

🛡️ Stability Systems

Atlas ships with a circuit breaker, load shedder, and emergency mode. If something goes wrong under load — runaway generation, memory pressure, tick overload — Atlas degrades gracefully instead of crashing or corrupting chunks.

---

Compatibility

Mod	Status
Lithium	✅ Fully compatible
Spark	✅ Fully compatible
Terralith	✅ Fully compatible
C2ME	⚠️ Not recommended — overlapping systems
Noisium	⚠️ Incompatible
Sodium / Iris	✅ Client-side, unaffected
ScalableLux	✅ Fully compatible(take with a grain of salt, not fully tested)

World generation datapacks that run on vanilla Minecraft are fully supported. Custom world generator mods generally work well. If you encounter a compatibility issue, please open an issue on GitHub.

---

FAQ

Does this change world generation output?
Atlas does not alter vanilla terrain, biomes, or structures. Due to the parallel nature of generation, chunk ordering is non-deterministic — worlds will vary run-to-run on the same seed. This is expected behavior, not a bug.

Is this server-side only?
Yes. Atlas only needs to be installed on the server (or in your singleplayer mods folder). Clients do not need Atlas installed to connect.

Does it work with Terralith and other datapacks?
Yes. In our benchmarks, Atlas with Terralith installed (5.69 cps) still outperformed C2ME running vanilla generation alone (4.84 cps).

What's the difference between the reported CPS and true CPS in your benchmarks?
Chunky's reported rate is a rolling window average that lags behind actual throughput. True CPS is calculated from total chunks divided by total elapsed time, which is the more accurate figure.

Should I use this with C2ME?
No. Both mods replace overlapping systems in the chunk pipeline and are not designed to run together.

---

Current Status

Atlas is currently in alpha. Core systems are stable and the mod has been benchmarked extensively, but it has not yet been tested across a wide range of hardware or mod configurations. Expect rough edges. Back up your worlds.

• ✅ DAG scheduler fully active
• ✅ PID throttle controller live-tuning
• ✅ Async I/O pipeline
• ✅ NUMA-aware thread assignment
• ⚠️ VULKAN I/O backend in progress (currently falls back to NIO2)
• 🔜 ScalableLux integrations
• 🔜 Broader mod compatibility testing

---

Atlas is made possible by countless hours of development and testing. If you'd like to support the project, consider leaving a star on GitHub or sharing your benchmark results.