deepstream-profile-pipeline

Profile DeepStream pipelines with Nsight Systems to derive optimal configuration parameters.

189|39|Updated May 8, 2026

Search Tags:#optimization #benchmarking #profiling #nsight-systems #video-analytics #deepstream

Ready to run

npx skills add https://github.com/NVIDIA/DeepStream --skill deepstream-profile-pipeline-nvidia

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: deepstream-profile-pipeline
Source: https://github.com/NVIDIA/DeepStream/tree/main/skills/deepstream-profile-pipeline
Command: npx skills add https://github.com/NVIDIA/DeepStream --skill deepstream-profile-pipeline-nvidia

SYSTEM DOCUMENTATION & REQUIREMENTS

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

What problem does it solve?

This skill eliminates the guesswork in DeepStream performance tuning by providing a data-driven, 6-stage measurement flow to identify bottlenecks and derive optimal configuration settings.

Core Features & Use Cases

Bottleneck Diagnosis: Automatically classifies pipelines as decode, compute, memory, or tracker-bound using Nsight Systems and nvidia-smi metrics.
Config Derivation: Applies closed-form rules to set optimal batch sizes, memory types, and inference parameters based on measured hardware ceilings.
Use Case: A developer struggling with low FPS on a multi-stream RTSP pipeline can use this skill to determine if they need a GPU with more NVDEC engines or a more powerful compute architecture, and receive the exact configuration changes required to reach peak performance.

Quick Start

Use the deepstream-profile-pipeline skill to profile the current pipeline and generate an optimized configuration report.

Frequently Asked Questions about deepstream-profile-pipeline

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

FAQPage Schema

How do I profile a DeepStream video analytics pipeline to find bottlenecks?▼

To profile a DeepStream pipeline, you can use Nsight Systems and nvidia-smi to measure hardware ceilings and automatically classify bottlenecks as decode, compute, memory, or tracker-bound. This data-driven approach eliminates guesswork in performance tuning.

Why does my multi-stream DeepStream pipeline have low FPS on NVIDIA hardware?▼

Low FPS in multi-stream DeepStream pipelines often stems from hitting decode, compute, memory, or tracker limits. By applying Nsight Systems profiling, you can identify whether you need more NVDEC engines or better compute architecture to reach peak performance.

Can I profile DeepStream performance tuning without a GUI on a headless server?▼

Yes, you can perform headless terminal-based profiling on DeepStream pipelines using nsys and nvidia-smi commands without any GUI dependencies. This allows you to derive optimal configuration parameters directly from remote dGPU or Jetson environments.

How do I determine the optimal batch size for multi-model inference in DeepStream?▼

To determine the optimal batch size for multi-model inference, apply closed-form rules based on measured hardware ceilings and inference micro-benchmarks gathered during Nsight Systems profiling. This process derives exact configuration changes required for peak performance.

What's the best way to optimize DeepStream configurations for both dGPU and Jetson platforms?▼

The best way to optimize DeepStream configurations for dGPU and Jetson platforms is running a 6-stage measurement flow using Nsight Systems. It diagnoses bottlenecks and applies closed-form rules to set optimal memory types and inference parameters from hardware ceilings.

Does Nsight Systems work with DeepStream to identify NVDEC engine limits?▼

Yes, Nsight Systems works with DeepStream to identify NVDEC engine limits by classifying pipelines as decode-bound using nvidia-smi metrics. It helps determine if a GPU with more NVDEC engines or a more powerful compute architecture is required.