GoKawiilDeveloper releases open-source agentic CUDA kernel optimizer built on LangGraph
A Show HN post introduces a tool that automatically generates, tests, benchmarks and refines CUDA GPU kernels from a workload description. It uses an LLM agent via LangGraph to propose code and launch-configuration changes, a C++/NVRTC harness to compile and run candidates, and Python to validate outputs against NumPy references, keeping only the fastest kernel that passes all correctness checks. The project was developed on Windows with an RTX 3060 Laptop GPU and requires Python 3.12+, CMake, a C++17 compiler and an OpenAI API key.
GoKawiil's interpretation of the reporting above, not reported fact.
Automating the write-test-profile-refine loop for GPU kernels could lower the expertise barrier for CUDA performance tuning, a task normally requiring deep hardware knowledge. Because ranking relies on geometric mean latency across benchmark cases and excludes compile and profiling overhead, the tool's outputs may not generalize to all real-world deployment scenarios. As an independent open-source project rather than an official NVIDIA release, its reliability and adoption remain unproven.
- Uses LangGraph-driven agent loop to generate, compile, test and benchmark CUDA kernels
- Combines a C++/NVRTC/CUDA Driver API harness with Python-based correctness checking
- Requires an OpenAI API key, CUDA Toolkit, CMake and a C++17 compiler to run
NVIDIA GeForce RTX 4070 GPU — If you're experimenting with CUDA kernel optimization like this agentic tool, having a capable local NVIDIA GPU makes iterative compiling, profiling with Nsight Compute, and benchmarking much faster. The RTX 4070 offers strong CUDA core counts and memory bandwidth for kernel development and testing without relying solely on cloud instances.
See NVIDIA GeForce RTX 4070 GPU on Amazon → Affiliate link — we may earn a commission on purchases, at no extra cost to you. Product picked by AI based on this article; it is not a tested recommendation.Source: github.com, 2026-09-25
Published there as: “Show HN: Agentic CUDA Kernel Optimizer”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.