Auto-research with codex: How I achieved a 232x Faster Kernel over baseline with Codex in GPU Mode's qr_v2 problem – sankalp's blog

Table of Contents Intro Contest in short Problem intro Why this problem is auto-research-able Learning Enough to Ask Better Questions (Optional) Math f...

08 Jul, 2026 Table of Contents Intro Contest in short Problem intro Why this problem is auto-research-able Learning Enough to Ask Better Questions (Optional) Math for QR decomposition: Householder reflections Make serial work small with the help of the blocked Householder algorithm Other challenges Codex-maxxing Kernel progress breakthroughs Breakthrough ideas Introducing idea diversity to escape the local maxima Implementation Hints What I could have done better Conclusion References Acknowledgements Intro Contest in short GPU Mode, in collab with Core Automation,…

saved by

• Benedict Neo
• Ishaan Panigrahi
• SUJASH AGRAWAL

related reading

• QR Decomp at the Speed of Lightml-mike.com
• GitHub - wafer-ai/gpu-perf-engineering-resources: A curated resource list for learning AI performance engineering, from GPU fundamentals to production inference.github.com
• Decoding Speculative Decoding from First Principlesjwlabs.vercel.app
• How to Optimize a CUDA Matmul Kernel for cuBLAS-like Performance: a Worklogsiboehm.com
• KernelBench: Can LLMs Write GPU Kernels?scalingintelligence.stanford.edu
• Modern GPU Programming For MLSys — Modern GPU Programming For MLSysmlc.ai
• Transformers Inference Optimization Toolset | AstraBlogastralord.github.io
• siboehmsiboehm.com
• Inside NVIDIA GPUs: Anatomy of high performance matmul kernels - Aleksa Gordićaleksagordic.com
• Kevin-32B: Multi-Turn RL for Writing CUDA Kernels | Cognitioncognition.ai
• About Us - Colfax Researchresearch.colfax-intl.com
• KernelBench v0.1 | Scaling Intelligence Lab at Stanford Universityscalingintelligence.stanford.edu