Aras PranckevičiusOy! @zeux talks about meshoptimizer on wookash podcast! https://www.youtube.com/watch?v=KtxiuFe83Hg
Great run down of all the topics there.
YouTube
@zeux Ren dropped by to hang out, but for some reason was *not* interested to learn about how vertex caches actually work.
@zeux btw Unity and mesh quantization: it actually has it, but done in arguably strange way. It had user configurable mesh quantization forever (like since 2010 or so), but that was only for *storage*, it was always uncompressed at load time (strange/stupid? yes).
Then later it *also* got "a bit" of actual quantization that stays quantized in memory, but with much less user control and more opaque when that would actually get used & how.
However, ...
@zeux ...it *does* allow you to create/modify meshes to be in fairly flexible attribute formats, so people who really want to do have decent quantization/formats (e.g. octahedral normals in sane amount of bits etc.), they can do it. But does not come out of the box. Why is a good question, "it's two dozen reasons that all compound" as always.
Topic jump: Arseny talks about meshoptimizer compression for *mesh* vertex data compression, but turns out, it is very good for lossless compression of other same-sized kinda-structured data. I've used it to compress game saves (https://aras-p.info/blog/2023/02/02/Float-Compression-4-Mesh-Optimizer/) and multilayer HDR images (https://aras-p.info/blog/2025/07/08/Lossless-Float-Image-Compression/) and it fared very well there. Very fast decompression speed too.
demofox@demofox yeah if you have an array of structs that tend to have spatially varying data, struct size is under 256B and is multiple of 4, then using meshopt vertex codec is very easy to try, and produces great results in my experience
Arseny Kapoulkine@aras She must be into splats instead of triangles