Codex自动优化CUDA内核232倍加速
推荐指数 77.0 NO. 007 · 2026.08.16
发布2026/08/15Score286Comments75
为什么值得看
GPU Mode编程竞赛中，参赛者用OpenAI Codex自动迭代优化QR分解CUDA内核，实现232倍加速获第12名。验证了AI辅助底层性能工程的可行性，对需要压榨硬件极限的AI Infra团队有参考价值。
编辑判断
这本质上是用LLM替代了传统性能工程师的试错循环，但关键不在'自动'而在反馈闭环设计——作者把profiler输出直接喂给Codex形成迭代，这比单纯让模型写代码有效得多。目前这类方法在矩阵分解这种结构清晰的数值计算上效果最好，图神经网络、稀疏计算等更复杂的场景还很难复现。
值得警惕的是'232x'这个数字的语境：baseline是未优化的纯PyTorch实现，而非cuSOLVER或CUTLASS等专业库，实际生产环境中边际增益会大幅收窄。做AI编译器或算子优化的团队可以借鉴其prompt工程思路，但别被竞赛数字误导投入产出比。
社区反馈
正面 71 条评论
核心争论：AI Agent能否自主完成底层性能优化闭环，实践者已验证可行
相关内容
In the last couple of days I wanted to try out the new definitive DeepSeek v4 releases. I gave it the repository of a semi-abandoned video compression codec and I told it to perform the usual benchmark -> profile -> verify -> research -> improve loop. I specifically chose this codec because the auth
I tried kernel autoreasearch using DeepSeek-V4-Flash as well. It spent about 1-2 hours to complete the FlashAttention optimization job (https://github.com/fengwang/FA5090/tree/main/v7) and cost me only $0.2. I believe we are ready to offload a lot of this kind well
> I tried kernel autoreasearch using DeepSeek-V4-Flash as well. It spent about 1-2 hours to complete the FlashAttention optimization job Doing the same, re-implementing a lot of LLM/diffusion models in Rust+CUDA for my own usage, usually the initial implementation takes 1-2 days (of 100% autono