QuantumHy is a server-side mod that makes your client run smoother in Hytale. It works by cutting how much the server tells your client to draw, and it adjusts that per player depending on where you are.
Active. The current development build targets small servers and solo worlds. Runtime behavior and config migration are covered by unit tests, but performance numbers below remain historical captures from earlier releases until the current build gets a new paired client capture.
- The server does not receive the client's FPS, so QuantumHy responds to entity density, streaming backlog, visible entity pressure, and server MSPT instead.
/q optimize offdisables every per-player lever for that player. Explicit server-wide admin controls such as world levers, spawn pause, and global LOD still affect the whole world.
The Hytale client is native, so no mod can touch the renderer. What a server mod can do is decide how much each client has to render, and that's the whole trick here: fewer chunks and entities in view means fewer things to draw, which means more FPS.
Two honest limits:
- It only helps where it's installed. Your singleplayer world, your own server, or a server that runs it. If you just join someone else's server, this can't do anything there.
- It never pushes your view further than you asked for. Your own view radius is the cap. QuantumHy only ever pulls it down, never up.
Every few seconds it checks how crowded the area around each player is (lots of entities means expensive to render) and turns that into one shrink factor for two levers: chunk view radius and entity stream radius.
- Packed area: both levers pull in toward the minimum. At full shrink the entity radius drops to the floor in one pass, not over many ticks.
- Out in the open: expand is allowed only after load has been calm for a couple of passes, at least half the density scan is loaded, you are not flying, and the world is not hitching. Radius never jumps from min to full in one pass (at most +1 chunk per check).
- Joining, streaming, or moving fast: it skips the density walk and holds expand so the client is not told to drop chunks it is still loading.
- World hitching (high MSPT): expand stays frozen and shrink can step faster until both the recent average and the last tick are calm.
Out of the box there's no hard cap. If you'd rather trade some view distance for FPS everywhere,
set targetClientViewRadius above 0.
It also smooths how fast chunks stream to you, so moving into fresh terrain arrives spread out
instead of in one burst that makes the client hitch. That's the smoothChunkStreaming keys below.
Each player can use /q optimize on|off. The preference is stored by UUID and does not change how
QuantumHy treats anyone else.
Client frametime capture on the Hytale Client process. Same test PC for every run below. Not lab benchmarks: different sessions, real play, spikes still happen.
| Component | Spec |
|---|---|
| CPU | AMD Ryzen 5 3600 |
| GPU | NVIDIA GeForce GTX 1650 4GB |
| RAM | 16 GB (2×8 GB) |
Solo world, same stress route (birds and mobs on screen). QuantumHy disabled vs default config enabled. Run lengths differ slightly (~205s off, ~233s on).
| Metric | Mod off | Mod on | Change |
|---|---|---|---|
| Average FPS | 102 | 146 | +43% |
| P5 | 51.6 | 91.1 | +76% |
| P1 | 36.2 | 64.4 | +78% |
| 1% low average | 14.0 | 19.8 | +41% |
| Time under 60 FPS | 25.3% | 6.9% | −73% of that slice |
P95 stayed high on both runs (~277 off, ~271 on). The big win is less time spent under 60 FPS and higher floor percentiles when the scene is busy. Massive frametime spikes (500ms–1s+) still show up on both runs, so QuantumHy helps sustained crowd load more than it removes one-off hitches.
| Metric | Mod off | Mod on |
|---|---|---|
| Recording length | ~205 s | ~233 s |
| Average FPS | 102 | 146.1 |
| P95 | 276.7 | 271.1 |
| P5 | 51.6 | 91.1 |
| P1 | 36.2 | 64.4 |
| 1% low average | 14.0 | 19.8 |
| P0.2 | 28.3 | 40.4 |
| P0.1 | 21.6 | 30.7 |
| 0.1% low average | 2.4 | 3.0 |
| Stuttering (time) | 5.3% (10.7 s) | 5.5% (12.8 s) |
| Smooth | 94.6% | 94.5% |
| Under 240 FPS | 94.4% | 92.4% |
| Under 60 FPS | 25.3% | 6.9% |
| Under 30 FPS | 5.5% | 4.8% |
| Under 10 FPS | 4.0% | 4.5% |
| Frametime variance < 2 ms | 70.0% | 86.5% |
Each capture is the full session: frametime graph on top, plus one analysis panel below.
Mod off
Mod on
Mod off: https://www.youtube.com/watch?v=H6Vns8b4hAg
Mod on: https://www.youtube.com/watch?v=CKGlXmX1M6k
Solo world, normal session (~5 min each). No intentional stress route, just regular play. QuantumHy disabled vs 0.2.1 default config enabled.
| Metric | Mod off | Mod on | Change |
|---|---|---|---|
| Average FPS | 159.6 | 226.9 | +42% |
| P5 | 111.1 | 178.1 | +60% |
| P1 | 77.2 | 139.1 | +80% |
| 1% low average | 43.7 | 111.7 | +155% |
| Time under 60 FPS | 2.2% | 0.1% | −95% of that slice |
The 0.2.1 recalibration is meant for real exploration, not only mob piles: higher average FPS, a much higher floor (P5/P1), less time under 60 FPS, and tighter frametime consistency (87.8% → 95.1% of frames with <2 ms variance). Chunk-load hitches still show up on both runs; the mod cuts how long you sit in them.
| Metric | Mod off | Mod on |
|---|---|---|
| Session length | ~5 min | ~5 min |
| Average FPS | 159.6 | 226.9 |
| P95 | 219.7 | 272.7 |
| P5 | 111.1 | 178.1 |
| P1 | 77.2 | 139.1 |
| 1% low average | 43.7 | 111.7 |
| P0.2 | 51.5 | 105.5 |
| P0.1 | 45.1 | 90.5 |
| 0.1% low average | 11.6 | 61.9 |
| Stuttering (time) | 1.9% | 0.2% |
| Smooth | 98.1% | 99.8% |
| Under 240 FPS | 99.5% | 64.3% |
| Under 60 FPS | 2.2% | 0.1% |
| Under 30 FPS | 1.2% | 0.1% |
| Under 10 FPS | 1.0% | 0.0% |
| Frametime variance < 2 ms | 87.8% | 95.1% |
Mod off
Mod on
Lives in QuantumHy.json in the plugin data folder, created on first run.
| Key | Default | What it does |
|---|---|---|
enabled |
true |
Turn the whole thing on or off. |
verboseLog |
false |
Log every pass with each player's density and view decision. Off on new installs. |
checkForUpdates |
true |
Notify operators or QuantumHy admins when the mod page lists a newer JAR. |
tickIntervalSeconds |
5 |
How often it re-checks each player. |
initialDelaySeconds |
20 |
Wait this long after start before the first pass. |
targetClientViewRadius |
0 |
Hard cap in chunks. 0 means no cap, just adapt. |
minClientViewRadius |
6 |
Never pull anyone below this. |
maxClientViewRadius |
32 |
Ceiling for the hard cap (your own view radius still wins). |
adaptiveTerrainViewEnabled |
true |
Adapt terrain view radius during normal play. |
emergencyTerrainTrimEnabled |
true |
Allow temporary terrain trim after sustained visual pressure. |
densityScanChunkRadius |
4 |
How many chunks around you it counts entities in. |
densityLowPerChunk |
1.0 |
Weighted entities per chunk at or below this: you get the full radius (minus optional baseline). |
densityHighPerChunk |
4.0 |
Weighted entities per chunk at or above this: you get pulled to the minimum. |
densityRingWeighting |
true |
Count center chunks fully, outer scan rings less (no per-species tables). |
densityRingEdgeWeight |
0.55 |
Ring weight at the scan edge when ring weighting is on (1.0 = flat count). |
baselineShrinkFraction |
0.10 |
Minimum shrink even in "open" density (0 = off). |
chunkLoadShrinkEnabled |
true |
Extra shrink from loaded + streaming section count (render backlog). |
chunkLoadLowChunks |
700 |
Loaded + loading sections at or below this: no chunk-load shrink. |
chunkLoadHighChunks |
1550 |
At or above this section count: chunk-load shrink hits full strength. |
densitySmoothing |
0.4 |
Smooths the density signal so a moving player's view doesn't flip-flop. Lower is smoother; 1.0 is off. |
adaptEntityRadius |
true |
Also shrink how far entities are streamed (not just chunks). The big win in mob-heavy spots. |
minEntityViewBlocks |
48 |
Never stream entities closer than this, in blocks (16 blocks = 1 chunk). |
entityLodAggressiveness |
1.0 |
Explicit global entity LOD override. 1.0 keeps the engine default. |
maxEntityVerticalDistance |
32 |
Drop entities too far above/below you from the stream (caves, ceilings). 0 = off. |
maxVisibleEntitiesPerPlayer |
80 |
Cap streamed entities per player in crowds (0 = off). |
holdSpawnOnLoadingChunks |
false |
Pause environmental spawning during a heavy client stream backlog. Global admin control. |
minViewRadiusDelta |
2 |
Don't bother changing the view for tiny differences. Ramped +1 expands still apply. |
maxExpandChunksPerPass |
1 |
Max chunks the client view radius may increase in one pass. |
maxShrinkChunksPerPass |
2 |
Max chunks it may decrease in one pass (doubled under MSPT pressure). |
maxExpandEntityBlocksPerPass |
16 |
Max entity-stream blocks the radius may increase in one pass. |
expandHysteresisPasses |
2 |
Calm passes required before an expand is allowed. |
worldPassBudgetMs |
8 |
Wall-clock budget for radius writes on the world thread. 0 disables the cap. |
pressureExitRequiresLastTick |
true |
Pressure only exits when both the 10s MSPT average and the last tick are calm. |
respectStreamingGrace |
true |
Don't shrink while you're still loading chunks. |
streamingBacklogThreshold |
80 |
How many loading sections counts as "still streaming". |
smoothChunkStreaming |
true |
Spread chunk streaming out so moving into new terrain doesn't hitch. |
maxChunksPerSecond |
128 |
Cap on sections streamed per second to a managed client. 0 keeps the engine default. |
maxChunksPerTick |
8 |
Cap on sections streamed per tick. 0 keeps the engine default. |
streamCatchUpEnabled |
true |
Raise the managed cap temporarily after fast travel while MSPT is healthy. |
streamCatchUpIntervalMs |
250 |
Interval for the lightweight stream controller. |
streamCatchUpPerSecond |
256 |
Temporary healthy catch-up cap per second. |
streamCatchUpPerTick |
12 |
Temporary healthy catch-up cap per tick. |
streamCatchUpHoldMs |
1500 |
Minimum time to hold catch-up after fast movement. |
leanCoreTakeover |
true |
If LeanCore is installed, take the view radius over from it (see below). |
yieldToLeanCoreViewRadius |
false |
The opposite: leave the view radius to LeanCore (see below). |
pressureGovernorEnabled |
true |
Tighten render levers when world MSPT stays high. |
pressureMsptEnter |
48 |
10s average MSPT at or above this enters pressure mode. |
pressureMsptExit |
43 |
MSPT at or below this exits pressure mode (hysteresis). |
pressureSustainSeconds |
4 |
How long MSPT must stay high before levers tighten. |
pressureCooldownSeconds |
15 |
How long MSPT must stay low before levers restore. |
pressureDensityMultiplier |
1.45 |
Under pressure, density thresholds tighten by this factor. |
pressureChunkRateMultiplier |
0.75 |
Under pressure, multiply chunk streaming caps. |
pressureLodMultiplier |
1.15 |
Pressure multiplier used only with an explicit global LOD value above 1.0. |
pressureVerticalTrimBlocks |
8 |
Under pressure, subtract from maxEntityVerticalDistance. |
pressureWorldLevers |
false |
Under pressure, pause NPC spawn and block tick on the world config (restored on release). |
pressureTrimClientEffects |
false |
Under pressure, trim bloom/sunshaft client effects (restored on release). |
pressureEffectScale |
0.5 |
Multiplier for client effect intensities while trimmed. |
Both mods can run together. Each one owns different levers, so they are not fighting over the
same ChunkTracker or Player fields.
| Lever | Who drives it (typical setup) |
|---|---|
| Client view radius | QuantumHy when leanCoreTakeover=true (default). QuantumHy turns off LeanCore view governance on startup. |
| Entity stream radius | QuantumHy |
| Hot/simulation radius | LeanCore |
Chunk send rate (maxChunks/s, maxChunks/tick) |
QuantumHy when smoothChunkStreaming=true (default) |
| MSPT render trim (density, LOD, vertical, effects) | QuantumHy |
| Spawn stream pause | QuantumHy |
| Zone dormancy / memory | LeanCore |
Defaults on both mods already line up: leanCoreTakeover=true, QuantumHy smoothChunkStreaming=true,
LeanCore handling sim/memory. No extra LeanCore config required for a normal install.
Knobs if you need them
leanCoreTakeover(defaulttrue): QuantumHy drives client view radius and disables LeanCore view governance. Setfalseonly if you want LeanCore on view radius instead (both mods may conflict).yieldToLeanCoreViewRadius(defaultfalse): QuantumHy stays out of view radius and chunk send-rate entirely. Use this if you deliberately want LeanCore in charge of those levers.
Do not flip on QuantumHy pressureWorldLevers alongside LeanCore unless you mean to pause NPC
spawn and block tick under MSPT. LeanCore does not touch those flags.
/q status shows live ownership (chunkRateOwner=, LeanCore: view=...) and the last adaptive
pass per player.
/q optimize on|off(aliases/quantumhy,/qhy): persistently enable or disable per-player optimization for yourself./q status: players see only their own mode and last decision. Admins and console see full server diagnostics./q help: lists the commands.
You need a Hytale install (that's where HytaleServer.jar comes from) and JDK 25.
./gradlew build
The jar lands in build/libs/. Drop it in %AppData%\Hytale\UserData\Mods\ on Windows.
MIT. See LICENSE.













