maxiaosong1124maxiaosong1124Communityยท1 Agent Skills Included

ncu-cuda-profiling-skill

Automated CUDA kernel profiling and bottleneck diagnosis

Automates NCU (Nsight Compute) profiling for CUDA kernels with full metrics collection and persistent report storage. Diagnoses five common bottleneck types like DRAM memory bound, L1 pressure, and low occupancy, then suggests concrete code fixes. Eliminates manual metric reading by generating ready-to-use Markdown and CSV performance reports with expected speedup estimates.
npx skills add maxiaosong1124/ncu-cuda-profiling-skill --all -g -y

All Skills in This Repository (1)

Pure Emerald Level Indicators

Frequently Asked Questions

FAQPage Schema
How to install ncu-cuda-profiling-skill?โ–ผ

Run `npx skills add maxiaosong1124/ncu-cuda-profiling-skill --all -g -y` in your terminal to install it globally for your AI coding tools.

How to find CUDA kernel performance bottlenecks?โ–ผ

Ask your AI assistant to analyze the kernel, and it runs NCU with full metrics collection, then classifies the bottleneck as memory bound, latency bound, compute bound, or occupancy bound with specific fixes.

What optimization suggestions does it provide?โ–ผ

It recommends concrete techniques like shared memory padding, block tiling, double buffering, and vectorized loads, each with expected speedup estimates ranging from 1.2x to 5x.

Does it work with Claude Code and Cursor?โ–ผ

Yes. It follows the standard SKILL.md format and works with Claude Code, Cursor, Codex, and Kimi Code CLI.

What are the system requirements?โ–ผ

You need Linux with CUDA 11.0 or later, Nsight Compute 2022.1 or later, and an NVIDIA Volta or newer GPU.

Related Repositories in Software Engineering

View All in Software Engineeringโ†’