deepstream-profile-pipeline — Claude Code skill: how to install
Profiles DeepStream pipelines with Nsight Systems to tune configs from benchmark data. Helps optimize batch sizes and locate bottlenecks without guessing.
npx skills add NVIDIA/skills --skill deepstream-profile-pipeline -a claude-codeWhat deepstream-profile-pipeline does
The skill configures and benchmarks video analytics pipelines running on DeepStream SDK 9.0. It starts by applying presets to nvinfer or nvinferserver, setting memory to NVMM, and selecting INT8 or FP16 precision. Afterwards, it triggers headless runs using nsys profile and nsys stats to collect execution metrics.
No graphical interface is needed. Profiling requires the nvcr.io/nvidia/deepstream:9.0-triton-multiarch development image on Ubuntu 22.04 or 24.04. The slim samples-multiarch container lacks the NVTX injector, which causes per-plugin traces to return empty data. Both nsys (version 2024+) and nvidia-smi must be accessible in your PATH.
What it helps with
- benchmark DeepStream pipelines using Nsight Systems
- tune batch sizes for nvinfer and nvstreammux
- detect decoder and inference bottlenecks with NVTX
- configure buffer queues and nvtracker presets
Example prompts for the agent
- Build an efficient DeepStream pipeline for 8 RTSP streams
- Profile this pipeline with nsys and locate the performance bottlenecks
- Optimize batch size and nvstreammux settings for maximum FPS
How to install deepstream-profile-pipeline
1Skills CLI (Vercel)
npx skills add NVIDIA/skills --skill deepstream-profile-pipeline -a claude-code2Manually: put the files in place
~/.claude/skills/deepstream-profile-pipeline/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/deepstream-profile-pipeline/
1Skills CLI (Vercel)
npx skills add NVIDIA/skills --skill deepstream-profile-pipeline -a codex2Manually: put the files in place
.agents/skills/deepstream-profile-pipeline/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add NVIDIA/skills --skill deepstream-profile-pipeline -a cursor2Manually: put the files in place
.cursor/skills/deepstream-profile-pipeline/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install NVIDIA/skills deepstream-profile-pipeline2Manually: put the files in place
.github/skills/deepstream-profile-pipeline/SKILL.md1Skills CLI (Vercel)
npx skills add NVIDIA/skills --skill deepstream-profile-pipeline -a gemini-cli2Platform CLI
gemini skills install https://github.com/NVIDIA/skills --path skills/deepstream-profile-pipelineManually: put the files in place
.devin/skills/deepstream-profile-pipeline/SKILL.mdThe legacy .windsurf/skills/ path still works
Skills CLI (Vercel)
npx skills add NVIDIA/skills --skill deepstream-profile-pipeline -a opencodeClaude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about deepstream-profile-pipeline
How to install deepstream-profile-pipeline in Claude Code?
Run npx skills add NVIDIA/skills --skill deepstream-profile-pipeline -a claude-code. The configuration is saved to ~/.claude/skills/deepstream-profile-pipeline/SKILL.md.
What does deepstream-profile-pipeline do?
It replaces guesswork by profiling DeepStream pipelines via Nsight Systems in the terminal. The skill determines batch limits, checks NVTX coverage, and adjusts element parameters to match target throughput.
Which Docker container is required?
Use the full dev container nvcr.io/nvidia/deepstream:9.0-triton-multiarch containing nsys and nvidia-smi. The samples-multiarch image will not work because its NVTX injector has been removed.
