Multi-Stream Video Analytics — DeepStream 9.1

DeepStream is NVIDIA's framework for building accelerated intelligent video-analytics (IVA) pipelines, and DeepStream 9.1 ships with JetPack 7.2 on Jetson Orin. This tutorial follows NVIDIA's official installation and quickstart documentation; every command below is taken from (or directly summarized from) those pages.

Version pairing: DeepStream 9.1 ↔ JetPack 7.2 GA ↔ L4T 39.2 ↔ CUDA 13.2 ↔ TensorRT 10.16.1.7 ↔ Ubuntu 24.04 ↔ GStreamer 1.24.2 (per NVIDIA's compatibility table).

1. Install

NVIDIA offers four install methods on Jetson; the official note recommends Docker for new users (fastest, dependency-free):

• Method 4 — Docker (recommended for new users): use the NGC DeepStream containers — see Docker Containers.
• Method 1 — SDK Manager: select DeepStreamSDK under "Additional SDKs" together with the JetPack 7.2 GA components.
• Method 2 — tar package: download deepstream_sdk_v9.1.0_jetson.tbz2 (from NVIDIA/DeepStream releases), then:

  sudo tar -xvf deepstream_sdk_v9.1.0_jetson.tbz2 -C /
  cd /opt/nvidia/deepstream/deepstream-9.1
  sudo ./install.sh
  sudo ldconfig

• Method 3 — Debian package: install deepstream-9.1_9.1.0-1_arm64.deb with sudo apt-get install ./deepstream-9.1_9.1.0-1_arm64.deb.

Prerequisite packages (official dependency list for the native install):

sudo apt install \
libssl3 libssl-dev libcurl4-openssl-dev \
libgstreamer1.0-0 gstreamer1.0-tools gstreamer1.0-plugins-good \
gstreamer1.0-plugins-bad gstreamer1.0-plugins-ugly gstreamer1.0-libav \
libgstreamer-plugins-base1.0-dev libgstrtspserver-1.0-0 \
libjansson4 libyaml-cpp-dev libmosquitto1

Juxi note: if you hit the documented RTSP issue (applications stuck at EOS with RTSP streams), run the update_rtpmanager.sh script in /opt/nvidia/deepstream/deepstream/ after installing the packages above.

2. Boost the clocks (before running anything)

sudo nvpmodel -m 0
sudo jetson_clocks

NVIDIA notes one exception: Jetson Orin Nano uses -m 2 for MAXN SUPER; all other Orin modules (including AGX Orin) use -m 0. Run these before running DeepStream applications.

3. Run the reference application

cd /opt/nvidia/deepstream/deepstream-9.1/samples/configs/deepstream-app
deepstream-app -c source30_1080p_dec_infer-resnet_tiled_display.txt

What to expect (per NVIDIA): a tiled display of 30 simulated 1080p streams with ResNet inference, and performance metrics — ~30 FPS for this configuration — printed in the terminal. Click a tile to zoom in; right-click to return to the tiled view.

Useful config files to explore (all in that directory):

Config	Use case
source30_1080p_dec_infer-resnet_tiled_display.txt	30-stream benchmark
source4_1080p_dec_infer-resnet_tracker_sgie_tiled_display.txt	Tracking + secondary inference
source1_usb_dec_infer_resnet.txt	Single USB camera
source1_csi_dec_infer_resnet.txt · source2_csi_usb_dec_infer_resnet.txt	CSI camera setups (driver support depends on your camera)
source2_1080p_dec_infer-resnet_demux.txt	Demux example

Notes from the official quickstart:

• First run with a new model takes minutes while the TensorRT engine is generated; later runs reuse it.
• If GStreamer elements fail to initialize, clear the cache: rm ${HOME}/.cache/gstreamer-1.0/registry.aarch64.bin
• Headless operation (no monitor): the default EGL sink needs a display. Configs support an RTSP output sink instead (see the [sink2] group in the 30-stream config) — stream the results to another machine.
• All precompiled sample apps live under /opt/nvidia/deepstream/deepstream-9.1/samples/ — each has a README.

4. What's new around DeepStream 9.1 on JetPack 7.2

• Agent-assisted pipelines: NVIDIA documents a DeepStream Coding Agent (AI agent support for building pipelines) — docs · GitHub.
• LLM/VLM in the pipeline: the reference apps include a deepstream-vllm-plugin for combining video pipelines with large-model reasoning — see the docs. For on-device model inference outside DeepStream, see Local LLM Inference.
• Triton on device: to run Triton Inference Server natively (without Docker), run sudo ./triton_backend_setup.sh in the samples directory (installs Triton 2.68.0 for Jetson).

Troubleshooting & further reading

• DeepStream Troubleshooting & FAQ
• Performance tuning — needed once you go beyond the reference configs
• Sample Configurations explained
• System-level issues (display, power, storage): see Troubleshooting

Sources

• DeepStream Installation Guide (checked 2026-09-24)
• DeepStream Quickstart Guide (checked 2026-09-24)
• JetPack 7.2.1 downloads page (checked 2026-09-24) — ⚠️ its component table lags on some rows; for the versions actually installed see Downloads

Status: draft, pending review by cheny. Grounded in NVIDIA's official documentation as of the date listed; not yet verified on physical hardware by Juxi Technology.

---

NVIDIA® and Jetson™ are trademarks of NVIDIA Corporation. This page is published by Juxi Technology and is not an NVIDIA publication.