Build & Release Pack / build (push) Successful in 1m39s

Details

feat: three-tier performance presets (Low/Medium/High) for client + server

Add drop-in performance presets tuned for three hardware tiers.

Tier targets:
- Low: 4GB VRAM/iGPU, 8-16GB RAM
- Medium: RX 580-class 8GB, 16-32GB RAM (shipped as pack default)
- High: RTX 3060+ 8GB+, 32GB+ RAM

Client configs (per tier, presets/<tier>/config/):
- sodium-options.json: chunk threads, cpu render-ahead, quality
- iris.properties: shaders on/off, max shadow render distance
- lambdynlights.toml: dynamic lighting mode + adaptive ticking
- entityculling.json: culling tracing distance
- cullleaves.json: leaf culling
- soundphysics.properties: reverb/ray budget (CPU cost)
- flywheel-client.toml: Create backend (BATCHING vs INSTANCING)

Server configs (per tier, presets/<tier>/server/):
- server.properties: view-distance, sim-distance, players, compression
- jvm-flags.txt: heap + G1 GC flags per RAM tier

Medium configs also copied to client_pack/config/ so they ship as the
mrpack defaults (bundled into overrides/config/).

Docs: docs/performance.md + docs/server.md rewritten for three tiers;
presets/README.md explains how to apply each tier.

Default (Medium) enables shaders with INSTANCING flywheel on the 8GB GPU
tier; Low disables shaders + uses BATCHING for maximum FPS on weak GPUs.

2026-08-19 23:27:45 -04:00

2.6 KiB

Raw Blame History

Server — Properties, JVM, Operations

Three server tiers are provided as drop-in presets in presets/<tier>/server/ (see presets/README.md). The Medium tier is the recommended default.

Tier	Host RAM	view-distance	simulation-distance	max-players
Low	8-16 GB	5	4	8
Medium	16-32 GB	8	6	8
High	32 GB+	12	10	16

Apply a tier (after packwiz install):

cp presets/medium/server/server.properties server_pack/server.properties
# then use the matching jvm-flags.txt when launching

server_pack/eula.txt → eula=true (after review).

Recommended JVM Flags (server, Medium)

java -Xms4G -Xmx8G \
  -XX:+UseG1GC \
  -XX:+ParallelRefProcEnabled \
  -XX:MaxGCPauseMillis=200 \
  -XX:+UnlockExperimentalVMOptions \
  -XX:+DisableExplicitGC \
  -XX:+AlwaysPreTouch \
  -XX:G1NewSizePercent=30 \
  -XX:G1MaxNewSizePercent=40 \
  -XX:G1HeapRegionSize=8M \
  -XX:G1ReservePercent=20 \
  -XX:G1HeapWastePercent=5 \
  -XX:G1MixedGCCountTarget=4 \
  -XX:InitiatingHeapOccupancyPercent=15 \
  -XX:G1MixedGCLiveThresholdPercent=90 \
  -XX:G1RSetUpdatingPauseTimePercent=5 \
  -Dlog4j2.formatMsgNoLookups=true \
  -jar neoforge-21.1.248-server.jar nogui

Notes:

G1 tuning keeps large-tech-server TPS stable under chunk gen (Krypton Reno + BadOptimizations help server-side).
-Xms4G avoids early GC churn. Cap at 8G for Medium; this pack is RAM-hungry from AE2/Mekanism tick load but Medium does not need 16G.
See presets/low/server/jvm-flags.txt (2-4G) and presets/high/server/jvm-flags.txt (8-16G) for the other tiers.

Server start order

packwiz install --dir server_pack (first time / after changes)
Accept EULA.
Launch with the JVM flags above.

Common failures

Symptom	Fix
"Mismatched mods" on client join	Run `packwiz refresh` both packs, redeploy.
Sinytra connector errors	Update Sinytra Connector + Forgified Fabric API together.
Iris black screen	Disable flywheel compat first, then test without shaders.
Crash on world gen (Deeper and Darker)	Ensure CF 659011 + its lithostitched data load; update both.
TPS lag from many AE2 cells	Check FTB chunk claims; lower `simulation-distance`.
Immersive Portals won't load	Must be the Fabric jar added via `url add --force` (Modrinth blocks loader mismatch).

Backups

World: copy server_pack/world/ (or use FTB Chunks backups).
Configs: server_pack/config/ mirrors client defaults; keep both in sync when changing KubeJS scripts or FTB settings.

2.6 KiB Raw Blame History