July 25, 2026
Fast physics, furious comments
SIMD for Collision
Game physics gets a huge speed boost, but fans are already fighting over who gets left out
TLDR: The Box3D creator showed a way to make heavy 3D collision checks much faster, especially in scenes packed with complex objects. Commenters were split between hype over the speed gains and frustration that ARM devices seem left out, with others joking that this “old” idea is suddenly being sold like brand-new magic.
A deeply nerdy performance post somehow turned into a mini comment-section soap opera. The core news is simple: the Box3D creator says he found a big way to make 3D collision checks much faster by handling several contact checks at once instead of one by one. That matters because when thousands of chunky 3D objects pile up, the amount of checking explodes and performance can fall off a cliff. His benchmark numbers suggest a massive speedup, which is the kind of thing engine developers absolutely live for.
But the comments? Oh, they instantly swerved into platform panic, old-tech skepticism, and SIMD confusion confessionals. The loudest complaint came fast: where’s the ARM support? One commenter basically called the whole thing “horrible” without it, turning a celebration post into a portability argument. Another reader played the wholesome student role, admitting they were a “SIMD noob” and saying the real revelation was that these tricks can help not just with math, but with moving data more efficiently too. That sparked the classic “wait, hasn’t this existed forever?” reaction, with one person asking why this suddenly gets treated like a miracle cure after decades in the shadows.
And because no internet thread can resist going off-script, someone dropped a podcast link with Erin Catto, while another just tossed in a mysterious YouTube link like a gremlin lobbing chaos into the room. So yes: faster physics, but the real collision was in the comments
Key Points
- •The article distinguishes wide SIMD, which processes multiple work units at once, from narrow SIMD, which packs vector components into SIMD registers.
- •A 3D convex pile benchmark with 5,120 convex hulls of 32 points each is used to illustrate why complex hull collision is much more expensive than box collision.
- •Box3D uses the Separating Axis Test for convex hull collision detection, while other engines may use GJK with EPA as an overlap fallback.
- •The article states that 3D SAT has quadratic complexity, with edge-edge combinations increasing from 144 for box-box tests to 7,921 for boulder-boulder tests.
- •Wide SIMD becomes worthwhile for large edge counts because one edge from one hull can be tested against four edges from another hull simultaneously, and benchmark results are reported from a 500-step run on an AMD 7950X.