2t-80t	1t-16t

Footnotes: https://gist.github.com/ishland/6eb0dd0af4216ffffd340ea994dc5796

2t-80t	1t-10t

Footnotes: https://gist.github.com/ishland/9601a033afad4be5b17ffaed859577fd

Benchmark setup

All benchmark done with single-socket Intel(R) Xeon(R) Platinum 8558P limited to 3.2GHz max, on Debian 13.

openjdk version "26.0.2" 2026-07-21
OpenJDK Runtime Environment Temurin-26.0.2+10 (build 26.0.2+10)
OpenJDK 64-Bit Server VM Temurin-26.0.2+10 (build 26.0.2+10, mixed mode)

Datapack used: terratonic-3.0.27.zip, Terralith_26.2_v2.6.4.zip

C2ME benchmark

Fabric mods used for C2ME benchmark:

c2me-fabric-mc26.2-0.4.2-alpha.0.41.jar
Chunky-Fabric-1.5.3.jar
fabric-api-0.157.0+26.2.jar
lithium-fabric-0.25.3+mc26.2.jar
ScalableLux-fabric-0.3.0-alpha.0.3-all.jar
spark-1.10.173-fabric.jar
vmp-fabric-mc26.2-0.2.0+beta.7.236-all.jar
zfastnoise-1.0.40+26.2.jar

Changed settings (c2me.toml):

• globalExecutorParallelism set to thread count

Paper benchmark

Paper used: paper-26.2-112.jar
Paper plugin used: Chunky-Bukkit-1.5.3.jar
Changed settings (paper-global.yml):

• worker-pool.io-threads = 4
• worker-pool.worker-threads set to thread count

Vanilla benchmark

Fabric mods used for vanilla benchmark:

Chunky-Fabric-1.5.3.jar
fabric-api-0.157.0+26.2.jar
spark-1.10.173-fabric.jar

Vanilla test uses a patched Chunky to force chunks to unload in order to avoid crashes during benchmark.
Extra JVM flags: -Dmax.bg.threads= set to thread count

All benchmarks are run on tmpfs, on overworld, with world folder emptied before each JVM execution.
Seed: 114862964781274189
force-load-existing-chunks is turned on in Chunky config.

2-96 threads benchmark

JVM flags: -Xmx32G -Xms32G -Xss4m -Xshare:off -XX:+DisableExplicitGC -XX:+UseNUMA -XX:+UseZGC -XX:-ZUncommit -XX:+UseLargePages -XX:+UseTransparentHugePages -XX:+UseCompactObjectHeaders -XX:+UnlockExperimentalVMOptions -XX:-OmitStackTraceInFastThrow -XX:+AlwaysPreTouch -XX:+UnlockDiagnosticVMOptions -XX:+DebugNonSafepoints -XX:+ExitOnOutOfMemoryError -Dchunky.maxWorkingCount=256
Warmup phase: 2048 block radius with 0 0 as center (data discarded)
Measurement phase: 6144 block radius with 1000000 1000000 as center (591361 chunks in total)

1-16 threads benchmark

JVM flags: -Xmx16G -Xms16G -Xss4m -Xshare:off -XX:+DisableExplicitGC -XX:+UseNUMA -XX:+UseZGC -XX:-ZUncommit -XX:+UseLargePages -XX:+UseTransparentHugePages -XX:+UseCompactObjectHeaders -XX:+UnlockExperimentalVMOptions -XX:-OmitStackTraceInFastThrow -XX:+AlwaysPreTouch -XX:+UnlockDiagnosticVMOptions -XX:+DebugNonSafepoints -XX:+ExitOnOutOfMemoryError -Dchunky.maxWorkingCount=256
Warmup phase: 1536 block radius with 0 0 as center (data discarded)
Measurement phase: 3072 block radius with 1000000 1000000 as center (148225 chunks in total)