Optimizing meshoptimizer to process billions of triangles in minutes (2025)

Let's be honest: most "performance" news in the graphics world is marketing fluff. "We made our engine 2x faster!" usually means they disabled shadows and called it a day. But Kapoulkine doesn't do fluff. meshoptimizer has been the quiet workhorse behind countless game pipelines for years — it's the tool that makes your Unity or Godot project actually ship on consoles with reasonable load times. When he says he's processing billions of triangles in minutes, that's real engineering, the kind that requires understanding cache hierarchies, SIMD instruction sets, and memory layout at a level most developers will never touch. What makes this particularly interesting is the open source angle. In a world where NVIDIA and Epic gatekeep their best tools behind NDAs and proprietary licenses, meshoptimizer remains MIT-licensed and available to anyone. This is the counterargument to the "open source can't compete on performance" crowd — here's a library that outperforms commercial alternatives while being completely transparent about how it works. You can read every line of code, understand every optimization, and contribute back if you're smart enough to keep up. The broader implication here is about the health of the graphics tooling ecosystem. We're seeing a renaissance of serious, performance-focused open source work — meshoptimizer, the Vulkan ecosystem, Blender's continued evolution — that's making high-end graphics development accessible to indies and hobbyists who would never get access to proprietary toolchains. Kapoulkine's work is a cornerstone of that movement. Every time he publishes one of these deep-dive posts, it's a masterclass not just in optimization, but in how to document engineering decisions for the public good. That said, I'd love to see some benchmarks against the commercial alternatives. The HN thread is suspiciously quiet — zero comments at time of writing — which either means nobody's had time to digest it or people are still benchmarking. Either way, if you're doing any serious mesh processing, you should be reading this article and thanking your lucky stars that this kind of expertise is being shared openly rather than locked behind a corporate firewall. This is how open source wins, one carefully optimized cache line at a time. Source: [https://zeux.io/2025/09/30/billions-of-triangles-in-minutes/](https://zeux.io/2025/09/30/billions-of-triangles-in-minutes/)