Very interesting to watch, though I don't really have a great idea of what's going on most of the time. Performance seems to be fairly poor despite my system being pretty beefy (Ryzen 9 7950x3d). I see the performance monitor and notice that the render loop seems to fairly regularly exceed the latency for 60fps, despite this being a 'simple' task by modern standards. I'd give more helpful feedback as to why, but the minified code makes it hard to say.
Do you plan on monetizing this somehow? If not, open sourcing some, if not all, would be pretty cool, even if it weren't necessarily licensed in a way that others could 'take' it, if that's your concern. Nonetheless, a very cool project.
FPS dips usually line up with big ecological events (population blooms, mass reproduction, lots of giant organisms, pathogen waves). The sim and rendering are decoupled, so you get (hopefully) brief spikes when a lot happens at once, then it settles again as populations crash/thin out. It’s normal behavior, not a runaway bug.
I’ve spent a lot of time optimizing hot paths and getting calculations down, but when the ecology goes a bit wild, you’ll still see temporary spikes. If you want to focus on stats only, you can switch to Simmer mode (no graphs) or turn off the arena, which reduces rendering work quite a bit.
Yes, some parts are inherently O(n²) (mate finding, crowd density, predator/prey proximity, pathogen spread). Ecology needs pairwise relationships.
To keep it sane, I don’t do naive all-vs-all. I use:
Zone-based spatial indexing so most checks only run against local neighbors (roughly n/16 instead of n). Temporal caching of indices so they’re not rebuilt every tick. Statistical sampling for crowd density at high population (estimate from a fixed-size sample instead of full scans).
So in practice it’s closer to O(n² / k), with k ≈ 16–50 depending on zone layout and population. You still see spikes during blooms, but it’s usually 10–30× faster than naive pairwise checks.
Could use quad trees, or similar bucketing. But I think he already stated that he tried to avoid the processing of long distance pairs. So probably he does some localized bucketing. Personally, I don't even fully understand what his simulation is doing. There are tradeoffs between accuracy / performance, depending on the specific problem. So hard to judge here.
Thanks for the feedback. There are some O(n²) components (mate finding, pathogen spread, proximity), but they’re mitigated with zone-based spatial hashing and throttling, so it’s not naive all-vs-all.
One big recent win was actually on the rendering side: I was rebuilding complex morphology geometry every frame for large organisms. I’ve added caching for that, which significantly reduced spikes during population blooms.
At very high populations the sim tick itself becomes the bottleneck, even for mostly O(n) logic per organism. When populations crash (which they usually do ecologically), performance recovers.
Quad trees / more adaptive structures are on my radar, but I’m trying to balance complexity vs predictability for now.
Do you plan on monetizing this somehow? If not, open sourcing some, if not all, would be pretty cool, even if it weren't necessarily licensed in a way that others could 'take' it, if that's your concern. Nonetheless, a very cool project.