Polycore 0.5.0 (Forge 1.21.11)
Curse Maven Snippet
What's new
- Minecraft 1.21.11 (Java 21): NeoForge (21.11.48 or later), Fabric (requires Fabric API) and Forge (61.2.5 or later).
- Fast aquifers (on by default): underground water and lava are decided for every air block during terrain
generation; Polycore now reuses the work shared by neighbouring blocks. Aquifers are about 25% faster and the
noise stage about 10% faster, with exactly the same world (checked chunk by chunk, blocks and scheduled fluids,
by an automated test). Switch: Optimizations page, or
/polycore toggle fast_aquifer. - Compatibility with other mods, checked at every start without knowing any mod by name: Polycore reads which Minecraft methods other mods change. An optimization whose code another mod also changes stays off; the menu (Optimizations, TPS Guard), the chest menu and the console say which mod and which method, and operators get a toast when they join, naming the mod, the method and its mixin class. It comes back by itself once that mod is removed. Item and XP orb merging are checked apart: with a mod that manages XP orbs, items still merge. The simulation distance measure also steps aside when something else changes the simulation distance. Polycore no longer replaces any method (the fast aquifers now add to vanilla's code), so other mods' changes to the same code keep working.
- HUD: asks the server only for the values it shows (about 0.2 KB instead of about 19 KB per refresh).
- TPS Guard, far mob AI: only the choice of goals and the brain sensors are slowed; running goals and behaviors tick exactly as in vanilla, so far mobs no longer stutter. Item merging keeps vanilla height.
- Menu: the "This server does not have Polycore" card grows with its text instead of letting the last line overflow in small windows.
- Menu: fits small windows (1280x720 at GUI scale 2): button labels are clipped, report lines wrap at words, tables shrink their columns instead of overflowing.
- Parity fingerprint:
/polycore hashnow reports a terrain fingerprint (relief, caves, aquifers, taken right after the carvers stage) next to the full one. Decorations spill into neighbouring chunks, so finished chunks depend on the order they were generated in; the terrain does not. Same seed and place always give the same terrain fingerprint, which is what proves an optimization changes nothing. Recording is on by default (settingparity.recordTerrain, measured at about 0.5 ms per new chunk on the generation threads, never on the server tick; off during benchmarks; the measured cost is written in each fingerprint report). - Reports page: a Copy button puts the open report in the clipboard. Fixed the Copy and Profiler Reset buttons being drawn just outside the visible area (they never showed).
- Removed the planned SIMD terrain noise: measured in Java 21, it was 3.5 to 4 times slower than vanilla (the noise table lookups do not vectorize), and vanilla's noise code is already well optimized by the JVM.
All Relations
- All Relations
- Embedded Library
- Optional Dependency
- Required Dependency
- Tool
- Incompatible
- Include
This mod has no related projects

