nsight

Profile GPU and CPU performance with timeline visualization and hardware counters.

4|Updated May 6, 2026
One-click install
npx skills add https://github.com/jstzwj/ai-infra-plugins --skill nsight
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: nsight
Source: https://github.com/jstzwj/ai-infra-plugins/tree/main/plugins/nsight/skills/nsight
Command: npx skills add https://github.com/jstzwj/ai-infra-plugins --skill nsight

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

This Skill provides comprehensive profiling and analysis of GPU workloads, CPU performance, and system events to optimize application efficiency.

Core Features & Use Cases

  • Timeline Visualization: Display detailed timelines of GPU activity, CPU backtraces, and API calls for performance bottleneck identification.
  • GPU Metrics & Hardware Profiling: Collect precise hardware counters like SM utilization, memory bandwidth, and power metrics.
  • Use Case: Developers analyzing deep learning training to pinpoint GPU idling or synchronization delays, improving throughput.
  • Multi-API Support: Trace Vulkan, DirectX, OpenGL, CUDA, OpenXR, and more to diagnose cross-API performance issues.

Quick Start

Run nsys profile with your application to generate an interactive report and identify bottlenecks.

Frequently Asked Questions about nsight

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

FAQPage Schema
How do I profile GPU and CPU performance to identify application bottlenecks?

To profile GPU and CPU performance, run `nsys profile` with your application to generate an interactive report. This timeline visualization displays GPU activity, CPU backtraces, and API calls to pinpoint synchronization delays.

What hardware counters are collected during GPU profiling?

GPU profiling collects precise hardware counters including SM utilization, memory bandwidth, and power metrics. These metrics help optimize graphics and compute workloads by revealing hardware-level performance constraints.

Can I trace Vulkan, DirectX, and CUDA API calls in the same profiling session?

Yes, you can trace Vulkan, DirectX, OpenGL, CUDA, and OpenXR API calls together. Multi-API support allows you to diagnose cross-API performance issues within a single system-wide profiling session.

Does GPU profiling require elevated permissions or specific driver versions?

Yes, NVIDIA GPU and CPU profiling requires compatible driver versions and elevated permissions. These prerequisites are necessary to access system events and hardware counters through various profiling APIs.

Why does my deep learning training experience GPU idling and synchronization delays?

GPU idling and synchronization delays during deep learning training occur when workloads are not optimally scheduled. Timeline visualization of GPU activity and CPU backtraces identifies exactly where these throughput bottlenecks happen.

What is the best way to analyze system-wide events and API traces for compute workloads?

The best way to analyze system-wide events is through timeline visualization and API trace analysis. This approach correlates GPU metrics with CPU performance data to optimize overall application efficiency.