Quickstart Guide¶

NVIDIA^® DeepStream Software Development Kit (SDK) is an accelerated AI framework to build intelligent video analytics (IVA) pipelines. DeepStream runs on NVIDIA^® T4, NVIDIA® Hopper, NVIDIA^® Ampere and platforms such as NVIDIA^® Jetson AGX Xavier™, NVIDIA^® Jetson Xavier NX™, NVIDIA^® Jetson AGX Orin™, NVIDIA^® Jetson Orin™ NX.

Note

NVAIE users should use dGPU/x86 Triton docker.

Jetson Setup¶

This section explains how to prepare a Jetson device before installing the DeepStream SDK.

Install Jetson SDK components¶

Download NVIDIA SDK Manager from https://developer.nvidia.com/embedded/jetpack. You will use this to install JetPack 5.1 GA (corresponding to L4T 35.2.1 release)

NVIDIA SDK Manager is a graphical application which flashes and installs the JetPack packages.
The flashing procedure takes approximately 10-30 minutes, depending on the host system.

Note

If you are using Jetson Xavier NX developer kit, you can download the SD card image from https://developer.nvidia.com/embedded/jetpack. This comes packaged with CUDA, TensorRT and cuDNN.

Install Dependencies¶

Enter the following commands to install the prerequisite packages:

$ sudo apt install \
libssl1.1 \
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

Install librdkafka (to enable Kafka protocol adaptor for message broker)¶

Clone the librdkafka repository from GitHub:

$ git clone https://github.com/edenhill/librdkafka.git

Configure and build the library:

$ cd librdkafka
$ git reset --hard 7101c2310341ab3f4675fc565f64f0967e135a6a
./configure
$ make
$ sudo make install

Copy the generated libraries to the deepstream directory:

$ sudo mkdir -p /opt/nvidia/deepstream/deepstream-6.2/lib
$ sudo cp /usr/local/lib/librdkafka* /opt/nvidia/deepstream/deepstream-6.2/lib

Install latest NVIDIA BSP packages¶

Installation of JetPack 5.1 GA will ensure that latest NVIDIA BSP packages are installed.

Install the DeepStream SDK¶

Method 1: Using SDK Manager

Select DeepStreamSDK from the Additional SDKs section along with JP 5.1 GA software components for installation.
Method 2: Using the DeepStream tar package: https://developer.nvidia.com/downloads/deepstream-sdk-v620-jetson-tbz2
1. Download the DeepStream 6.2 Jetson tar package deepstream_sdk_v6.2.0_jetson.tbz2 to the Jetson device.
2. Enter the following commands to extract and install the DeepStream SDK:
  $ sudo tar -xvf deepstream_sdk_v6.2.0_jetson.tbz2 -C / $ cd /opt/nvidia/deepstream/deepstream-6.2 $ sudo ./install.sh $ sudo ldconfig
Method 3: Using the DeepStream Debian package: https://developer.nvidia.com/downloads/deepstream-62-620-1-arm64-deb

Download the DeepStream 6.2 Jetson Debian package deepstream-6.2_6.2.0-1_arm64.deb to the Jetson device. Enter the following command:
```
$ sudo apt-get install ./deepstream-6.2_6.2.0-1_arm64.deb
```
Method 4: Use Docker containers

DeepStream docker containers are available on NGC. See the Docker Containers section to learn about developing and deploying DeepStream using docker containers.

Run deepstream-app (the reference application)¶

Navigate to the samples directory on the development kit.
Enter the following command to run the reference application:
```
$ deepstream-app -c <path_to_config_file>
```
Where <path_to_config_file> is the pathname of one of the reference application’s configuration files, found in configs/deepstream-app/. See Package Contents for a list of the available files.

Config files that can be run with deepstream-app:
1. source30_1080p_dec_infer-resnet_tiled_display_int8.txt
2. source30_1080p_dec_preprocess_infer-resnet_tiled_display_int8.txt
3. source4_1080p_dec_infer-resnet_tracker_sgie_tiled_display_int8.txt
4. source4_1080p_dec_infer-resnet_tracker_sgie_tiled_display_int8_gpu1.txt (dGPU only)
5. source1_usb_dec_infer_resnet_int8.txt
6. source1_csi_dec_infer_resnet_int8.txt (Jetson only)
7. source2_csi_usb_dec_infer_resnet_int8.txt (Jetson only)
8. source6_csi_dec_infer_resnet_int8.txt (Jetson only)
9. source2_1080p_dec_infer-resnet_demux_int8.txt
10. source4_1080p_dec_infer-resnet_tracker_sgie_tiled_display_int8.yml
11. source30_1080p_dec_infer-resnet_tiled_display_int8.yml
12. source4_1080p_dec_preprocess_infer-resnet_preprocess_sgie_tiled_display_int8.txt
Note
- You can find sample configuration files under /opt/nvidia/deepstream/deepstream-6.2/samples directory. Enter this command to see application usage:
  $ deepstream-app --help
- To save TensorRT Engine/Plan file, run the following command:
  $ sudo deepstream-app -c <path_to_config_file>

To show labels in 2D Tiled display view, expand the source of interest with mouse left-click on the source. To return to the tiled display, right-click anywhere in the window.

Keyboard selection of source is also supported. On the console where application is running, press the z key followed by the desired row index (0 to 9), then the column index (0 to 9) to expand the source. To restore 2D Tiled display view, press z again.

Boost the clocks¶

After you have installed DeepStream SDK, run these commands on the Jetson device to boost the clocks:

$ sudo nvpmodel -m 0
$ sudo jetson_clocks

Note

For Jetson Xavier NX, run sudo nvpmodel -m 8 instead of 0.

Run precompiled sample applications¶

Navigate to the chosen application directory inside sources/apps/sample_apps.
Follow the directory’s README file to run the application.

Note

If the application encounters errors and cannot create Gst elements, remove the GStreamer cache, then try again. To remove the GStreamer cache, enter this command: $ rm ${HOME}/.cache/gstreamer-1.0/registry.aarch64.bin

When the application is run for a model which does not have an existing engine file, it may take up to a few minutes (depending on the platform and the model) for the file generation and the application launch. For later runs, these generated engine files can be reused for faster loading.

dGPU Setup for Ubuntu¶

This section explains how to prepare an Ubuntu x86_64 system with NVIDIA dGPU devices before installing the DeepStream SDK.

Note

This document uses the term dGPU (“discrete GPU”) to refer to NVIDIA GPU expansion card products such as NVIDIA Tesla^® T4 , NVIDIA GeForce^® GTX 1080, NVIDIA GeForce^® RTX 2080 and NVIDIA GeForce^® RTX 3080. This version of DeepStream SDK runs on specific dGPU products on x86_64 platforms supported by NVIDIA driver 525.85.12 and NVIDIA TensorRT™ 8.5.2.2 and later versions.

You must install the following components:

Ubuntu 20.04
GStreamer 1.16.3
NVIDIA driver 525.85.12
CUDA 11.8
TensorRT 8.5.2.2

Remove all previous DeepStream installations¶

Enter the following commands to remove all previous DeepStream 3.0 or prior installations:

$ sudo rm -rf /usr/local/deepstream /usr/lib/x86_64-linux-gnu/gstreamer-1.0/libgstnv* /usr/bin/deepstream* /usr/lib/x86_64-linux-gnu/gstreamer-1.0/libnvdsgst*
/usr/lib/x86_64-linux-gnu/gstreamer-1.0/deepstream*
/opt/nvidia/deepstream/deepstream*
$ sudo rm -rf /usr/lib/x86_64-linux-gnu/libv41/plugins/libcuvidv4l2_plugin.so

To remove DeepStream 4.0 or later installations:

Open the uninstall.sh file in /opt/nvidia/deepstream/deepstream/
Set PREV_DS_VER as 4.0
Run the following script as sudo ./uninstall.sh

Install Dependencies¶

Enter the following commands to install the necessary packages before installing the DeepStream SDK:

$ sudo apt install \
libssl1.1 \
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 \
libjsoncpp-dev \
protobuf-compiler \
gcc \
make \
git \
python3

Install CUDA Toolkit 11.8¶

Run the following commands (reference, https://developer.nvidia.com/cuda-downloads):

$ sudo apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/3bf863cc.pub
$ sudo add-apt-repository "deb https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/ /"
$ sudo apt-get update
$ sudo apt-get install cuda-toolkit-11-8

Note

If you observe following errors while CUDA installation, refer to https://developer.nvidia.com/blog/updating-the-cuda-linux-gpg-repository-key/.

W: GPG error: https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64 InRelease: The following signatures couldn't be verified because the public key is not available: NO_PUBKEY A4B469963BF863CC
E: The repository 'https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64 InRelease' is no longer signed.

Install NVIDIA driver 525.85.12¶

Download and install using NVIDIA driver 525.85.12 from NVIDIA Unix drivers page at: https://www.nvidia.com/Download/driverResults.aspx/198879/en-us/

Run the following commands:

$chmod 755 NVIDIA-Linux-x86_64-525.85.12.run
$sudo ./NVIDIA-Linux-x86_64-525.85.12.run --no-cc-version-check

Note

Ensure gdm, lightdm or Xorg service is stopped while installing nvidia driver

Use command : sudo service gdm stop sudo service lightdm stop sudo pkill -9 Xorg

Install TensorRT 8.5.2.2¶

Run the following command to install TensorRT 8.5.2.2:

sudo apt-get install libnvinfer8=8.5.2-1+cuda11.8 libnvinfer-plugin8=8.5.2-1+cuda11.8 libnvparsers8=8.5.2-1+cuda11.8 \
libnvonnxparsers8=8.5.2-1+cuda11.8 libnvinfer-bin=8.5.2-1+cuda11.8 libnvinfer-dev=8.5.2-1+cuda11.8 \
libnvinfer-plugin-dev=8.5.2-1+cuda11.8 libnvparsers-dev=8.5.2-1+cuda11.8 libnvonnxparsers-dev=8.5.2-1+cuda11.8 \
libnvinfer-samples=8.5.2-1+cuda11.8 libcudnn8=8.7.0.84-1+cuda11.8 libcudnn8-dev=8.7.0.84-1+cuda11.8 \
python3-libnvinfer=8.5.2-1+cuda11.8 python3-libnvinfer-dev=8.5.2-1+cuda11.8

Note

It is assumed, “deb https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/ /” has been added, as mentioned in CUDA Toolkit 11.8 installation step.

Install librdkafka (to enable Kafka protocol adaptor for message broker)¶

Clone the librdkafka repository from GitHub:

$ git clone https://github.com/edenhill/librdkafka.git

Configure and build the library:

$ cd librdkafka
$ git reset --hard 7101c2310341ab3f4675fc565f64f0967e135a6a
./configure
$ make
$ sudo make install

Copy the generated libraries to the deepstream directory:

$ sudo mkdir -p /opt/nvidia/deepstream/deepstream-6.2/lib
$ sudo cp /usr/local/lib/librdkafka* /opt/nvidia/deepstream/deepstream-6.2/lib

Install the DeepStream SDK¶

Method 1: Using the DeepStream Debian package

Download the DeepStream 6.2 dGPU Debian package deepstream-6.2_6.2.0-1_amd64.deb : https://developer.nvidia.com/downloads/deepstream-62-620-1-amd64-deb

Enter the command:
```
$ sudo apt-get install ./deepstream-6.2_6.2.0-1_amd64.deb
```
Method 2: Download the DeepStream tar package: https://developer.nvidia.com/downloads/deepstream-sdk-v620-x86-64-tbz2

Navigate to the location of the downloaded DeepStream package to extract and install the DeepStream SDK:
```
$ sudo tar -xvf deepstream_sdk_v6.2.0_x86_64.tbz2 -C /
$ cd /opt/nvidia/deepstream/deepstream-6.2/
$ sudo ./install.sh
$ sudo ldconfig
```
Method 3: Use Docker containers

DeepStream docker containers are available on NGC. See the Docker Containers section to learn about developing and deploying DeepStream using docker containers.

Run the deepstream-app (the reference application)¶

Go to the samples directory and enter this command:
```
$ deepstream-app -c <path_to_config_file>
```
Where <path_to_config_file> is the pathname of one of the reference application’s configuration files, found in configs/deepstream-app. See Package Contents for a list of the available files.
Note
- To dump engine file, run the following command:
  
  $ sudo deepstream-app -c <path_to_config_file>
- You can find sample configuration files under /opt/nvidia/deepstream/deepstream-6.2/samples directory. Enter this command to see application usage:
  
  $ deepstream-app --help
To show labels in 2D tiled display view, expand the source of interest with a mouse left-click on the source. To return to the tiled display, right-click anywhere in the window.
Keyboard selection of source is also supported. On the console where application is running, press the z key followed by the desired row index (0 to 9), then the column index (0 to 9) to expand the source. To restore the 2D Tiled display view, press z again.

Run precompiled sample applications¶

Navigate to the chosen application directory inside sources/apps/sample_apps.
Follow that directory’s README file to run the application.
Note

If the application encounters errors and cannot create Gst elements, remove the GStreamer cache, then try again. To remove the GStreamer cache, enter this command:
$ rm ${HOME}/.cache/gstreamer-1.0/registry.x86_64.bin
When the application is run for a model which does not have an existing engine file, it may take up to a few minutes (depending on the platform and the model) for the file generation and application launch. For later runs, these generated engine files can be reused for faster loading.

dGPU Setup for RedHat Enterprise Linux (RHEL)¶

Deepstream for RHEL is not supported in this release.

How to visualize the output if the display is not attached to the system¶

1 . Running with an X server by creating virtual display¶

Refer https://docs.nvidia.com/grid/latest/grid-vgpu-user-guide/index.html#configuring-xorg-server-on-linux-server for details.

2 . Running without an X server (applicable for applications supporting RTSP streaming output)¶

The default configuration files provided with the SDK have the EGL based nveglglessink as the default renderer (indicated by type=2 in the [sink] groups). The renderer requires a running X server and fails without one. In case of absence of an X server, DeepStream reference applications provide an alternate functionality of streaming the output over RTSP. This can be enabled by adding an RTSP out sink group in the configuration file. Refer to [sink2] group in source30_1080p_dec_infer-resnet_tiled_display_int8.txt file for an example. Don’t forget to disable the nveglglessink renderer by setting enable=0 for the corresponding sink group.

Platform and OS Compatibility¶

The following table provides information about platform and operating system compatibility in the current and earlier versions of DeepStream.

Jetson model Platform and OS Compatibility¶

DS release

DS 6.0

DS 6.0.1

DS 6.1

DS 6.1.1

DS 6.2

Jetson platforms

Nano, AGX Xavier, TX2, TX1, Jetson NX

Nano, AGX Xavier, TX2, TX1, Jetson NX

AGX Xavier, Jetson NX, Jetson Orin

AGX Xavier, Jetson NX, Jetson Orin

Xavier AGX , Xavier NX, AGX Orin, Orin NX

OS

L4T Ubuntu 18.04

L4T Ubuntu 18.04

L4T Ubuntu 20.04

L4T Ubuntu 20.04

L4T Ubuntu 20.04

JetPack release

4.6 GA

4.6.1 GA

5.0.1 DP

5.0.2 GA Revision 1

5.1 GA

L4T release

32.6.1

32.7.1

34.1.1

35.1

35.2.1

CUDA release

CUDA 10.2

CUDA 10.2

CUDA 11.4

CUDA 11.4

CUDA 11.4

cuDNN release

cuDNN 8.2.1.32

cuDNN 8.2.1.32

cuDNN 8.3.2.49

cuDNN 8.4.1.50+

cuDNN 8.6.0.166+

TensorRT release

TRT 8.0.1

TRT 8.2.1

TRT 8.4.0.11

TRT 8.4.1.5

TRT 8.5.2.2

OpenCV release

OpenCV 4.1.1

OpenCV 4.1.1

OpenCV 4.2.0

OpenCV 4.2.0

OpenCV 4.2.0

VisionWorks

VisionWorks 1.6.502

VisionWorks 1.6.502

NA

NA

NA

GStreamer

GStreamer 1.14.5

GStreamer 1.14.5

GStreamer 1.16.2

GStreamer 1.16.2

GStreamer 1.16.3

Docker image

deepstream-l4t:6.0-ga

deepstream-l4t:6.0.1

deepstream-l4t:6.1

deepstream-l4t:6.1.1

deepstream-l4t:6.2

dGPU model Platform and OS Compatibility¶

DS release

DS 6.0

DS 6.0.1

DS 6.1

DS 6.1.1

DS 6.2

GPU platforms

P4, T4, V100, A100

P4, T4, V100, A100

T4, V100, A2, A10, A30, A100, RTX Ampere (Ax000/RTX30x0)

T4, V100, A2, A10, A30, A100, RTX Ampere (Ax000/RTX30x0)

T4, V100, A2, A10, A30, A100, RTX Ampere (Ax000/RTX30x0), Hopper, ADA

OS

Ubuntu 18.04 RHEL 8.x

Ubuntu 18.04 RHEL 8.x

Ubuntu 20.04

Ubuntu 20.04

Ubuntu 20.04

GCC

GCC 7.3.0

GCC 7.3.0

GCC 9.4.0

GCC 9.4.0

GCC 9.4.0

CUDA release

CUDA 11.4.1

CUDA 11.4.1

CUDA 11.6.1

CUDA 11.7.1

CUDA 11.8

cuDNN release

cuDNN 8.2+

cuDNN 8.2+

cuDNN 8.4.0.27

cuDNN 8.4.1.50+

cuDNN 8.7.0.84-1+

TRT release

TRT 8.0.1

TRT 8.0.1

TRT 8.2.5.1

TRT 8.4.1.5

TRT 8.5.2.2

Display Driver

R470.63.01

R470.63.01

R510.47.03

R515.65.01

R525.85.12

VideoSDK release

SDK 9.1

SDK 9.1

SDK 9.1

SDK 9.1

SDK 9.1

OFSDK release

2.0.23

2.0.23

2.0.23

2.0.23

2.0.23

GStreamer release

GStreamer 1.14.5

GStreamer 1.14.5

GStreamer 1.16.2

GStreamer 1.16.2

GStreamer 1.16.3

OpenCV release

OpenCV 3.4.0

OpenCV 3.4.0

OpenCV 4.2.0

OpenCV 4.2.0

OpenCV 4.2.0

Docker image

deepstream:6.0-ga

deepstream:6.0.1

deepstream:6.1

deepstream:6.1.1

deepstream:6.2

Note

By default, OpenCV has been deprecated. However, OpenCV can be enabled in plugins such as nvinfer (nvdsinfer) and dsexample (gst-dsexample) by setting WITH_OPENCV=1 in the Makefile of these components. Please refer component README for more instructions.

DeepStream Triton Inference Server Usage Guidelines¶

To migrate the Triton version in a DeepStream 6.2 deployment (Triton 22.09) to a newer version (say Triton 22.10 or newer), follow the instructions at DeepStream Triton Migration Guide.

dGPU¶

Pull the DeepStream Triton Inference Server docker

docker pull nvcr.io/nvidia/deepstream:6.2-triton

Start the docker

docker run --gpus "device=0" -it --rm -v /tmp/.X11-unix:/tmp/.X11-unix -e DISPLAY=$DISPLAY -e CUDA_CACHE_DISABLE=0 nvcr.io/nvidia/deepstream:6.2-triton

Note

The triton docker for x86 is based on tritonserver 22.09 docker and Jetson is based on tritonserver 23.01 docker, and has Ubuntu 20.04.
When the triton docker is launched for the first time, it might take a few minutes to start since it has to generate compute cache.

Jetson¶

DeepStream Triton container image (nvcr.io/nvidia/deepstream-l4t:6.2-triton) has Triton Inference Server and supported backend libraries pre-installed.

In order to run the Triton Inference Server directly on device, i.e., without docker, Triton Server setup will be required.

Go to samples directory and run the following commands to set up the Triton Server and backends.

$ cd /opt/nvidia/deepstream/deepstream/samples/
$ sudo ./triton_backend_setup.sh

Note

By default script will download the Triton Server version 2.30. For setting up any other version change the package path accordingly.

Triton backends are installed into /opt/nvidia/deepstream/deepstream/lib/triton_backends by default by the script. User can update infer_config settings for specific folders as follows:

model_repo {
  backend_dir: /opt/nvidia/tritonserver/backends/
}

Using DLA for inference¶

DLA is Deep Learning Accelerator present on the Jetson AGX Xavier, Jetson NX platforms, Jetson AGX Orin and Jetson Orin NX. These platforms have two DLA engines. DeepStream can be configured to run inference on either of the DLA engines through the Gst-nvinfer plugin. One instance of Gst-nvinfer plugin and thus a single instance of a model can be configured to be executed on a single DLA engine or the GPU. However, multiple Gst-nvinfer plugin instances can be configured to use the same DLA. To configure Gst-nvinfer to use the DLA engine for inference, modify the corresponding property in nvinfer component configuration file (example: samples/configs/deepstream-app/config_infer_primary.txt): Set enable-dla=1 in [property] group. Set use-dla-core=0 or use-dla-core=1 depending on the DLA engine to use.

DeepStream does support inferencing using GPU and DLAs in parallel. You can run this in separate processes or single process. You will need three separate sets of configs configured to run on GPU, DLA0 and DLA1:

Separate processes¶

When GPU and DLA are run in separate processes, set the environment variable CUDA_DEVICE_MAX_CONNECTIONS as 1 from the terminal where DLA config is running.

Single process¶

DeepStream reference application supports multiple configs in the same process. To run DLA and GPU in same process, set environment variable CUDA_DEVICE_MAX_CONNECTIONS as 32:

$ deepstream-app -c <gpuconfig> -c <dla0config> -c <dla1config>