Skip to content
Programming

One x86 instruction takes 62 seconds and the leaderboard is glorious

hackernews / top 1d ago 5

The Assembly Hall of Shame is a GitHub repository that inverts the usual goal of instruction latency analysis: instead of finding the fastest instructions, it hunts for the absolute slowest single x86 instruction. The current champion is fxrstor64, which took 62 seconds (198 billion cycles) on an AMD Ryzen 7 5800H by loading 512 bytes of FPU state from a high-latency MMIO region while other cores saturated the PCIe fabric with competing non-posted transactions. The repo includes a full leaderboard ranking instructions from nop (1 cycle) up to the champion, with creative strategies for maximizing each instruction's latency.

Read full article →

More Programming

AI Loops Can Run Forever — Too Bad That's the Problem

tldr / product 1d ago 7

AI agents can loop forever, pass every visible test, and still miss the user's intent entirely — the real challenge is teaching them when to quit, NBD.

AI ate the apprenticeship; taste is the last thing left on the plate

tldr / webdev 1d ago 7

AI collapsed the idea-to-artifact distance to near zero, leaving taste as the only remaining differentiator; the person who spent years making bad code to learn judgement now ships at the same speed as someone who skipped the apprenticeship entirely, NBD.

Your Kubernetes Controller's Cache: The Silent RAM Buffet Between You and the API Server

tldr / devops 1d ago 7

Turns out r.Get() doesn't hit the API server — it reads from a local cache, which is great until your controller quietly eats gigabytes of RAM doing O(n) scans nobody asked for, NBD.