Same decisions, sooner.
A Timberborn 1.1 mod for big colonies that stutter. It removes work the game repeats over and over, so you get fewer lag spikes and shorter freezes.
Played in late-game colonies of 350+ beavers, most recently as 0.4.30. This release hasn't been played yet. Needs Harmony and Mod Settings; try it on a copy of a save. What is tested, and what isn't
Every hauler decision, finished road and lumberjack's search makes the game redo work whose answer mostly hasn't changed. In a small colony that costs nothing. With hundreds of beavers it adds up, and you feel it as stutter.
This mod keeps answers the game has already worked out, does heavy work on other CPU cores, and skips lookups that can't change the result. Beavers make the same choices, just sooner.
Six jobs the game repeats all the time, each done faster and the game's own way. Most can be checked against the game while you play.
For every hauler looking for work, the game asks every building in the district for its jobs and sorts the list. The mod keeps each building's jobs until something changes, then sorts the list the game's way.
Late Game PerformanceHaulCacheVerifyAfter a path, stair or platform is finished, the game rebuilds every building's route map on the main thread. The mod spreads that over up to 7 worker threads. The maps come out exactly as the game builds them.
To find a tree, the game measures the path to every marked tree, though most are still growing. The mod leaves out the measurements that can't change the answer, and a lumberjack with no grown tree in reach stops at the first tree. Lumberjacks, gatherers and farmers get the game's own result.
YielderSearchVerifyTrees, crops, paths, levees and platforms, nearly all of a big colony, are saved on worker threads. Beavers and workplaces stay on the main thread. The save file is the same.
SaveSnapshotVerifyThe game cleans up memory with everything stopped. In a big colony that can freeze it for most of a second, about once a minute. One tick-box spreads the same work over many frames.
After a long frame, such as an autosave, the game crams what it missed into the next few frames. In a logged 354-beaver session, each hitch became about a second of stutter. The mod runs the same ticks, in the same order, over ordinary frames.
[LateGamePerformance]Simulation features:Player.log. It reads the same on every computer running this version.In multiplayer every player's game must calculate exactly the same thing. A mod that changed even one answer would be worse than no mod. So keeping co-op games in step comes before every speed-up.
Incremental garbage collection, thread counts, timing logs and the verify settings. They don't affect the simulation, so each player can choose them. Everything else needs the same mod version and the same game version, updated together.
Every figure says where it was measured and what it doesn't promise. Your gain depends on your colony and your computer. There is no controlled frame-rate benchmark yet.
Test harness, not a live game: a model forest of 2,000 marked trees, 50 of them grown. Fewer is better.
| Case | Lookups |
|---|---|
| Game's own search | 2,000 |
| With the mod | 51 |
How much it saves in your colony depends on your forest: a lot when most marked trees are still growing, nothing when every marked tree is grown.
Test harness: a 22,600-tile road network, heavier than a real colony's. Less is better.
| Case | Time |
|---|---|
| One at a time | about 1.8 s |
| On 7 workers | about 0.23 s |
In the game the workers run alongside the tick. The main thread still builds a map itself if it needs one before a worker gets to it.
Played: a 362-beaver colony with Verify every feature on, so the game's code ran beside the mod's and every answer was compared.
| Part of the tick | Game's own | With the mod | Answers |
|---|---|---|---|
| Hauler job list, per request | 0.94 ms | 0.24 ms | Same · 0 differences |
| District resource counts | 0.68 ms | 0.56 ms | Same · 0 differences |
Tree search, plant water, water map copy, soil scans, terrain search, home search and reachability also reported 0 differences in that session. The overall tick time did not measurably drop in a session that short.
Played: the colony at 362 beavers, against a recording of it at 359 beavers made on an earlier build of this mod. So these show what the newer work added, not the whole gain over the unmodded game.
| Measure | Earlier build | Now | Change |
|---|---|---|---|
| Autosave freeze | 0.84–1.26 s | 325–430 ms | Shorter |
| Entities saved on the main thread | 3,670 | 411 | Fewer |
| Plant water, per tick | 0.31 ms | 0.03 ms | Faster |
| Animations, per frame | 1.64 ms | 0.87 ms | Faster |
| Beaver decisions, per tick | 9.3 ms | 8.6 ms | Faster |
One session each, one computer. The overall tick time did not measurably drop in that session.
Played: one 35-minute multiplayer session, same save and same mods, on two computers.
| Without incremental collection | With it | |
|---|---|---|
| Freezes from memory clean-up | 39, median 675 ms each (about 30 s in total) | 1 frame over 50 ms outside saves |
| In play | Drastic spikes at high tick rates | A much steadier frame rate |
Two different computers, so hardware counts as well as the setting. Memory in use was 1.6 to 2.7 GB on both.
Test harness, run against the installed game's own assemblies.
Tested extensively, but still a young mod, so here is the honest picture. Try it on a copy of a save first.
This release, 0.4.31, hasn't been played yet. Everything under In play is from earlier builds.
Documents\Timberborn\Mods.