March 14, 2026
CPU glow-up, GPU side-eye
A Recursive Algorithm to Render Signed Distance Fields
Old CPUs are suddenly drawing wild 3D — and the comments are losing it
TLDR: A clever CPU-only method draws complex 3D scenes surprisingly fast, hitting 30–100 FPS for some users and ~250 FPS with optimizations. The community is split between “CPU renaissance” hype and “neat demo, still impractical” skepticism, but everyone agrees it’s a flashy step that could simplify development
The graphics nerds just dropped a brain-bender: a new way to draw 3D worlds with mathy “distance” tricks — and it runs shockingly well on plain old CPUs. The thread exploded with dueling vibes: one camp swooning over demoscene wizardry and tiny 64KB demos, the other chanting “cool, but still slow.” The vibe-setter? A commenter who summed it up as “SDFs are really slow but cool”, then bragged the author’s CPU-only method hits about 50 frames per second on one core, and even 100 FPS by reusing patches of pixels. Cue the mic drop from another user: “Holy crap! 30 FPS on my decade-old CPU.”
Then a speed freak showed up, sprinkled in some vector magic (SIMD), and claimed ~250 FPS avg on an M4 chip — instantly birthing the meme: “Bring GPU tricks back to the CPU.” The hot debate: is this the start of a CPU comeback, or just a gorgeous party trick? Optimizers say threading is “straightforward,” skeptics say “try a real scene.” Jokes flew about “Lisp wizards” vs “Fortran dads,” and one user dubbed it “functional art school for triangles.” Love it or doubt it, the crowd agrees on this: it’s wild, it’s nerdy, and it’s fast enough to start fights
Key Points
- •SDFs define 3D objects via functions that return signed distance, enabling powerful composition and procedural scenes.
- •Inigo Quilez helped popularize SDFs; demoscene productions (e.g., by Mercury) showcased their potential in 64KB demos.
- •Raymarching (sphere tracing) is the common SDF rendering method; it’s simple but often more computationally expensive than rasterization or ray tracing.
- •Prior optimization ideas include partial evaluation and interval arithmetic, though they may require compiler-like infrastructure; bounding volumes can help cull work.
- •The author proposes a CPU-oriented recursive divide-and-conquer approach that subdivides the view frustum and advances rays together to reduce SDF rendering cost.