KernelWiki

Official

Optimize NVIDIA GPU kernels with expert knowledge

Authormit-han-lab
Version1.0.0
Installs0

System Documentation

What problem does it solve?

This Skill provides expert knowledge for optimizing NVIDIA Blackwell (SM100, B200) and Hopper (SM90, H100) GPU kernels, enabling users to improve performance and efficiency.

Core Features & Use Cases

  • Kernel Optimization: Offers insights into performance patterns, techniques, and specific kernel case studies for both architectures.
  • Performance Patterns: Helps identify issues like low SM utilization, memory-bound, register pressure, and compute-bound scenarios.
  • DSLs & Languages: Provides guidance on using CuTe DSL, CUDA C++ with PTX inline, Triton on Blackwell, and other relevant languages/DSLs.
  • Migration Guides: Offers guidance for migrating Hopper features to Blackwell.
  • PR References: Allows users to query specific PRs for implementation details.

Quick Start

Use the KernelWiki skill to search for information on optimizing a specific kernel, such as python3 scripts/query.py "optimization for FlashAttention-4".

Dependency Matrix

Required Modules

PyYAML

Components

scriptsreferencesassets

💻 Claude Code Installation

Recommended: Let Claude install automatically. Simply copy and paste the text below to Claude Code.

Please help me install this Skill:
Name: KernelWiki
Download link: https://github.com/mit-han-lab/KernelWiki/archive/main.zip#kernelwiki

Please download this .zip file, extract it, and install it in the .claude/skills/ directory.
View Source Repository

Agent Skills Search Helper

Install a tiny helper to your Agent, search and equip skill from 620,000+ vetted skills library on demand.