profile-analyzer

Analyze NVIDIA Nsight Systems profiles for GPU performance and regressions.

1.0k|109|Updated Jun 19, 2024
One-click install
npx skills add https://github.com/sirius-db/sirius --skill profile-analyzer
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: profile-analyzer
Source: https://github.com/sirius-db/sirius/tree/main/.claude/skills/profile-analyzer
Command: npx skills add https://github.com/sirius-db/sirius --skill profile-analyzer

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) and assets (resource) components.

What problem does it solve?

This Skill helps developers and engineers deeply analyze the performance of GPU-accelerated applications by processing NVIDIA Nsight Systems (nsys) profiles, identifying bottlenecks, and generating comprehensive reports.

Core Features & Use Cases

  • Comprehensive GPU Analysis: Analyzes kernel occupancy, memory bandwidth, operator attribution, and more from nsys profiles.
  • Performance Regression Detection: Compares different profiling runs to identify performance regressions or improvements.
  • Workflow Guidance: Provides clear workflows for profiling, timing, and comparison, distinguishing between profiled (for analysis) and non-profiled (for timing) runs.
  • Use Case: A developer has made changes to a GPU-accelerated SQL query engine and wants to understand the impact on performance. They can use this Skill to generate detailed reports on kernel execution, memory transfers, and operator efficiency to pinpoint any performance regressions introduced by their changes.

Quick Start

Use the profile-analyzer skill to perform a full performance analysis on Sirius GPU profiles by running the nsys_report.sh script with a scale factor of 1.

Frequently Asked Questions about profile-analyzer

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I analyze NVIDIA Nsight Systems profiles to identify GPU bottlenecks?

To analyze NVIDIA Nsight Systems (nsys) profiles and identify GPU bottlenecks, use this skill to process existing profiles and generate detailed reports on kernel occupancy, memory bandwidth, and operator attribution.

What is the best way to detect performance regressions in GPU-accelerated applications?

Detecting performance regressions in GPU-accelerated applications is achieved by comparing multiple nsys profiling runs. This skill generates comparative reports highlighting changes in kernel execution and memory transfers to pinpoint regressions or improvements.

Can I generate a GPU performance report from an existing nsys profile?

Yes, you can generate a GPU performance report from an existing nsys profile. This skill supports workflows specifically for report generation from existing profiles, analyzing kernel occupancy and memory bandwidth without requiring a new profiling run.

Does GPU performance analysis distinguish between profiled and non-profiled runs?

GPU performance analysis distinguishes between profiled runs for detailed analysis and non-profiled runs for timing. This skill provides clear workflow guidance to ensure you use the correct execution mode for accurate performance regression detection.

How do I perform a quick GPU performance analysis on nsys profiles?

To perform a quick GPU performance analysis on nsys profiles, use the quick analysis workflow provided by this skill. It efficiently processes nsys data to deliver immediate insights into kernel execution and memory bandwidth.