Show HN: Agentic CUDA Kernel Optimizer

Category: devtools

Tags: cuda, gpu-optimization, llm-agents, langgraph, kernel-generation

Score: 6.5/10 (Innovation: 7, Technical: 7, Documentation: 7, Utility: 5)

An agentic CUDA kernel optimizer that uses LangGraph and LLM-driven code generation to iteratively propose, compile, validate, and benchmark CUDA kernels, feeding results back for refinement. It integrates NVRTC compilation, Nsight Compute profiling, and NVIDIA documentation research to guide optimization. The project is interesting for its combination of LLM agents with GPU performance tuning, though it's early-stage and experimental.

Target audience: GPU developers, performance engineers, AI/ML infrastructure engineers

Repository: https://github.com/bertaye/agentic-cuda-optimizer · Python · 37 stars

View on Hacker News