Smooth Server

Smooth Server brings together a small selection of mods aimed at improving server tick processing, chunk handling and memory use. This NeoForge edition targets Minecraft 26.1.2 and is designed for dedicated servers. Players do not need to install this modpack to join.

Alpha release: this edition is still being evaluated. It has passed local multiplayer and save/restart tests, but broader compatibility and long-term stability need more testing. Back up your world and configuration before trying it on an existing server.

Included mods

Mod	Version	Purpose
Moonrise	1.0.1+2cd8eef	Chunk handling and server optimizations
Lithium	0.24.7	Game logic and tick optimizations
FerriteCore	9.0.0	Memory optimizations
Chunky	1.5.4	Optional world pregeneration
spark	1.10.187	Performance profiling and diagnostics
No Chat Reports	26.1-v2.19.0	Changes chat-reporting behavior

Chunky and spark provide tools you can run when needed. Pregeneration and profiling do not start automatically.

Measured performance

We compared six NeoForge configurations in three matched runs per configuration, using separate copies of the same test worlds and running one server at a time. The workloads covered chunk generation and loading, two-player exploration, mob AI, tracking, collisions, hoppers, furnaces and redstone.

Baseline versus the initial optimized stack

The table below compares NeoForge with common benchmark tools (A) against Moonrise + Lithium + FerriteCore + ServerCore (D). Both used the same tools and JVM settings. D is an earlier comparison build: the current alpha removes ServerCore and adds No Chat Reports. These figures must not be read as a direct benchmark of the downloadable alpha.

Measurement	NeoForge baseline A	Initial optimized stack D
Chunk generation throughput	24.50 chunks/s	27.41 chunks/s
Mixed synthetic workload, two players — mean MSPT	8.17 ms	4.60 ms
Mob pathfinding, tracking and collisions — mean MSPT	4.18 ms	2.45 ms
Hoppers and furnaces — mean MSPT	1.73 ms	0.95 ms
Clocked redstone — mean MSPT	4.84 ms	2.67 ms
Fresh-terrain exploration, two players — mean / p99 MSPT	6.16 / 46.24 ms	3.92 / 33.90 ms
Pregenerated-terrain exploration — mean / p99 MSPT	8.33 / 52.13 ms	5.57 / 54.64 ms
Fresh-terrain exploration — sampled peak process RAM	1,361 MiB	1,577 MiB

Values are averages across three runs; p99 values are averages of each run's individual-tick 99th percentile. Lower MSPT is better; higher chunk throughput is better. TPS stayed near 20 in most phases, so lower tick times indicate additional headroom rather than a demonstrated increase in player capacity. Generation measured a 1,089-chunk Overworld selection after a separate warmup.

The optimized stack did not improve every metric. It used more process RAM in this comparison, and its pregenerated-terrain p99 tick time was slightly worse despite a lower mean. Individual runs varied; these are local measurements, not guaranteed gains for every server.

Configuration comparison, before the final alpha was assembled. A = baseline; B = Moonrise; C = C2ME; D = Moonrise + Lithium + FerriteCore + ServerCore; E = C2ME + Lithium + FerriteCore + ServerCore; F = Lithium + FerriteCore + ServerCore. All include common measurement tools. Generation error bars show run standard deviation. C/E's successful generation results do not establish reliability: separate startup/spawn failures prevented their selection.

How the final stack was selected

Separate matched tests removed Lithium, FerriteCore and ServerCore individually. The selected stack keeps Lithium and FerriteCore and removes ServerCore to reduce known vanilla-behavior differences. Removing ServerCore produced similar generation throughput in that experiment: 26.11 versus 26.52 chunks/s for its matched control. Fresh-terrain mean / p99 MSPT was 4.80 / 45.28 ms versus 5.10 / 51.35 ms, but paired tick results varied enough that an improvement is inconclusive. Separate decoded chunk-loading tests also showed a modest loading cost from removing ServerCore. These tests are a different experiment from the A–D table above and should not be combined into a claimed baseline-to-final speedup.

Each variant removes one mod from D. “Without ServerCore” is the selected performance stack before No Chat Reports was added. Error bars show variation across runs; the lower average p99 is inconclusive because paired outcomes varied.

The selected Moonrise + Lithium + FerriteCore + Chunky + spark stack then completed three separate 20-minute sustained runs with two scripted clients, fresh-terrain excursions, saves, reconnects and active fixtures:

Measurement	Observed range across the three runs
Measured TPS	19.65–19.71
Mean MSPT	5.71–5.94 ms
p99 MSPT	34.56–38.01 ms
Sampled peak process RAM	1,885–1,911 MiB
Largest recorded GC pause	71.9–72.7 ms

All three runs shut down cleanly and passed fixture checks, saved-region scans and live world reloads. Process RAM rose during exploration; these 20-minute runs do not establish whether memory use eventually plateaus.

Three runs of the selected five-mod performance stack. Left: sampled process RAM over wall time, including activity between measured cycles. Right: p99 tick time for each measured two-minute cycle. The table reports percentiles across each complete measured run, so its values differ from individual cycle percentiles. No Chat Reports was absent from these runs.

No Chat Reports was added after those performance tests. The current six-mod alpha passed a shorter two-client activity, save and restart test. A matched baseline comparison and sustained qualification of this exact downloadable alpha have not yet been completed.

Test conditions and limits

Tests used Minecraft 26.1.2, NeoForge 26.1.2.116, Temurin Java 25.0.4+7, two physical CPU cores, a 2 GiB Java heap and a 4 GiB Java-process memory limit on one Windows PC. View and simulation distances were both 6. Scripted clients connected over loopback. Every comparison arm used the same calibrated G1 settings, including disabled concurrent refinement; these are not measurements of an untouched default JVM configuration.

The tests combine synthetic fixtures with scripted multiplayer activity. They do not establish performance on other hardware, large public servers or overnight workloads. Cold Windows filesystem-cache loading, WAN behavior and exhaustive vanilla world-generation parity were not tested. Performance varies with the world, player activity and any additional mods.

Gameplay and compatibility

The pack adds no new blocks, mobs or progression systems. It aims to stay close to vanilla, with known technical differences: Moonrise retains its default five-tick catch-up limit, and No Chat Reports changes chat-reporting behavior. Exact vanilla behavior is not guaranteed for every technical build or farm.

Adding content or other optimization mods creates a different configuration. Test additions before using them on an important world, and include logs and reproduction steps when reporting problems.

Installation

1. Use Java 25 and a server installer that supports .mrpack files and NeoForge.
2. Install the file matching Minecraft 26.1.2. This alpha pins NeoForge 26.1.2.116.
3. Review the Minecraft EULA and server settings before starting.
4. Adjust the supplied small-server defaults to your needs: two players, view distance 6 and simulation distance 6. Review the JVM settings for your available resources.

This is a server-only pack. Install it through a compatible server tool rather than treating it as a client performance pack.

All credit for the included mods belongs to their authors. Visit the linked projects for documentation, source code and support.