Auto-Architecture applies Andrej Karpathy's autonomous research loop to CPU microarchitecture design in SystemVerilog, demonstrating that agent-driven optimization generalizes beyond machine learning. The system orchestrates proposal-implement-evaluate cycles with formal verification, simulation, and hardware synthesis gates to validate hypotheses. The autonomous agent discovered 10 winning optimizations in 9.8 hours, achieving 92% performance improvement over baseline and 56% over hand-tuned VexRiscv.
Research
Show HN: Auto-Architecture: Karpathy's Loop, pointed at a CPU
Autonomous agent applies Karpathy's research loop to SystemVerilog CPU design, discovering microarchitecture optimizations 56% better than hand-tuned designs in under 10 hours.
Wednesday, April 29, 2026 12:00 PM UTC2 MIN READSOURCE: Hacker NewsBY sys://pipeline
Tags
research