NVIDIA Jetson AGX Thor - H264 GStreamer Pipelines

Follow Us On

Share This Page

Facebook Twitter LinkedIn WhatsApp Email Copy Link

Close

NVIDIA®Jetson AGX Thor™
Introduction
SoM Overview Carrier Boards NVIDIA developer kit Auvidea X241 Connect Tech Gauntlet
JetPack 7.0
Getting Started Installing JetPack Flashing the board with GUI installer JetPack Components Compiling Kernel and DTB Flashing the Board from Cmdline Flashing Filesystem Flashing the Board from USB Performance Tuning Evaluating Performance Set Values Manually Tuning Power Maximizing Performance
Development in the Board
Getting into the Board Using Serial Console Using SSH Using Ubuntu Login Installing Packages
Camera Support
Camera Over Ethernet (CoE) MIPI CSI IMX477+J20 Driver
Blackwell GPU
GStreamer Pipelines
Capture and Display H264 H265
Holoscan
PVA
RidgeRun Products
GstCUDA GstRtspSink GstStitcher GstNvStabilize GstBEV GstQtOverlay GstRrBrowserSink RidgeRun Development Suite Spherical Video PTZ
RidgeRun Open Source Projects
GstPerf GStreamer Daemon GstInterpipe GstShark
RidgeRun Reference Designs
Panoramic Stitching and WebRTC Streaming
Reference Documentation
Contact Us

GStreamer Pipelines using H264

This section presents GStreamer pipelines to capture from MIPI CSI-2 and USB cameras and encoding/decoding of video using the H.264 hardware codec of the AGX Thor. Key performance indicators are measured using two power profiles (operation modes). Check our GStreamer Pipelines section to find more information about how we extracted the performance metrics presented in this section.

H264 Encoding

The following table compare the utility and describes each available encoder in the Jetson AGX Thor for H.264 encoding.

Table 1: H.264 Encoders comparison

Feature	GStreamer plugin
	nvh264enc	nvcudah264enc	nvautogpuh264enc	nvv4l2h264enc
Backend	NVENC API	NVCODEC API CUDA Mode	NVCODEC API auto GPU select Mode	V4L2 API
GPU Handling	Fixed to GPU0 (read-only)	Manual (cuda-device-id)	Auto GPU (override possible)	gpu-id property (writable)
Memory types	CUDA, GLMemory, system	CUDA, GLMemory, system	CUDA, GLMemory, system	NVMM only
Input formats	NV12, YV12, I420, BGRA, RGBA, Y444, VUYA (widest)	NV12, Y444 only	NV12, Y444 only	I420, NV12, P010_10LE, Y444, Y444_10LE, NV24
Output caps	H.264 (byte-stream, AU aligned)	H.264 (avc/byte-stream, AU aligned)	H.264 (avc/byte-stream, AU aligned)	H.264 (byte-stream, AU or NAL aligned)
B-frames	0–4	0–4	0–4	0–2 (num-B-Frames)
Special features	zerolatency, strict-gop	multi-pass, repeat-sequence-header, cc-insert, zero-reorder-delay	multi-pass, repeat-sequence-header, cc-insert, zero-reorder-delay	Intra-refresh, CABAC toggle, SEI UUID, per-frame latency measure

MIPI CSI-2 Camera

The pipelines in this section used the Framos IMX676 Sensor . For reference the testing used the following variables.

FILE=/tmp/filename.mp4
WIDTH=3552
HEIGHT=3556
FRAMERATE=64/1
FORMAT=NV12

Encoding with nvv4l2h264enc and saving to file

The pipeline below can be used to capture, encode using the H.264 hardware codec, and store the video to file as an mp4 in the /tmp/filename.mp4 location. Table 1 shows the performance information for this pipeline.

gst-launch-1.0 nvarguscamerasrc video/x-raw, width=$WIDTH, height=$HEIGHT, framerate=$FRAMERATE ! nvvidconv ! 'video/x-raw(memory:NVMM), format=$FORMAT' ! nvv4l2h264enc ! h264parse ! qtmux ! filesink location=$FILE -e

Table 1: Performance of pipeline with nvv4l2h264enc

Operation Mode	CPU (%)	GPU 1 (%)	GPU 2 (%)	FPS	latency (ms)
0 (max performance)
1 (Default Mode)

Encoding with nvautogpuh264enc and saving to file

The pipeline below can be used to capture, encode using the H.264 hardware codec, and store the video to file as an mp4 in the /tmp/filename.mp4 location. Table 2 shows the performance information for this pipeline.

gst-launch-1.0 nvarguscamerasrc video/x-raw, width=$WIDTH, height=$HEIGHT, framerate=$FRAMERATE ! nvvidconv ! 'video/x-raw(memory:NVMM), format=$FORMAT' ! nvv4l2h264enc ! h264parse ! qtmux ! filesink location=$FILE -e

Table 2: Performance of pipeline with nvv4l2h264enc

Operation Mode	CPU (%)	GPU 1 (%)	GPU 2 (%)	FPS	latency (ms)
0 (max performance)
1 (Default Mode)

Encoding with nvh264enc and saving to file

The pipeline below can be used to capture, encode using the H.264 hardware codec, and store the video to file as an mp4 in the /tmp/filename.mp4 location. Table 3 shows the performance information for this pipeline.

gst-launch-1.0 nvarguscamerasrc video/x-raw, width=$WIDTH, height=$HEIGHT, framerate=$FRAMERATE ! nvvidconv ! 'video/x-raw(memory:NVMM), format=$FORMAT' ! nvv4l2h264enc ! h264parse ! qtmux ! filesink location=$FILE -e

Table 3: Performance of pipeline with nvv4l2h264enc

Operation Mode	CPU (%)	GPU 1 (%)	GPU 2 (%)	FPS	latency (ms)
0 (max performance)
1 (Default Mode)

USB Camera

The pipelines in this section used the LI-10635 USB camera.

For reference the testing used the following variables.

FILE=/tmp/filename.mp4
WIDTH=1920
HEIGHT=1080
FRAMERATE=30/1

Encoding nvv4l2h264enc and saving to file

The pipeline below can be used to capture, encode using the h.264 hardware codec, and store the video to file as an mp4 in the /tmp/filename.mp4 location. Table 4 shows the performance information for this pipeline.

gst-launch-1.0 v4l2src device=/dev/video2 ! video/x-h264, width=$WIDTH, height=$HEIGHT, framerate=$FRAMERATE ! h264parse ! video/x-h264, stream-format=avc ! h264parse ! video/x-h264, stream-format=byte-stream ! nvv4l2decoder ! queue ! nvvidconv ! 'video/x-raw(memory:NVMM)' ! nvv4l2h264enc bitrate=8000000 ! video/x-h264, stream-format=byte-stream ! h264parse ! qtmux ! filesink location=$FILE -e

Table 4: Performance of pipeline with nvv4l2h264enc

Operation Mode	CPU%	GPU1%	GPU2%	FPS
0 (max performance)
1 (Default Mode)

Encoding nvautogpuh264enc/nvcudah264enc and saving to file

The pipeline below can be used to capture, encode using the h.264 hardware codec, and store the video to file as an mp4 in the /tmp/filename.mp4 location. Table 5 shows the performance information for this pipeline.

gst-launch-1.0 v4l2src device=/dev/video2 ! video/x-h264, width=$WIDTH, height=$HEIGHT, framerate=$FRAMERATE  ! h264parse ! video/x-h264, stream-format=avc ! h264parse ! video/x-h264, stream-format=byte-stream ! queue max-size-buffers=2 leaky=2 ! nvh264dec ! queue max-size-buffers=2 leaky=2 ! nvautogpuh264enc ! video/x-h264, stream-format=byte-stream ! h264parse ! qtmux ! filesink location=/tmp/filename.mp4 -e

Table 5: Performance of pipeline with nvv4l2h264enc

Operation Mode	CPU%	GPU1%	GPU2%	FPS
0 (max performance)
1 (Default Mode)

Encoding nvh264enc and saving to file

The pipeline below can be used to capture, encode using the h.264 hardware codec, and store the video to file as an mp4 in the /tmp/filename.mp4 location. Table 6 shows the performance information for this pipeline.

gst-launch-1.0 v4l2src device=/dev/video2 ! video/x-h264, width=$WIDTH, height=$HEIGHT, framerate=$FRAMERATE  ! h264parse ! video/x-h264, stream-format=avc ! h264parse ! video/x-h264, stream-format=byte-stream ! queue max-size-buffers=2 leaky=2 ! nvh264dec ! queue max-size-buffers=2 leaky=2 ! nvh264enc ! video/x-h264, stream-format=byte-stream ! h264parse ! qtmux ! filesink location=/tmp/filename.mp4

Table 6: Performance of pipeline with nvv4l2h264enc

Operation Mode	CPU%	GPU1%	GPU2%	FPS
0 (max performance)
1 (Default Mode)

H264 Decoding

The pipeline below can be used to decode the videos recorded with both the MIPI CSI-2 and the USB cameras.

FILE=/tmp/filename.mp4
gst-launch-1.0 filesrc location=$FILE ! qtdemux ! queue ! h264parse ! nvv4l2decoder ! nv3dsink -e

Table 7: Performance of the pipeline decoding video

Operation Mode	CPU%	GPU1%	GPU2%	FPS
0 (max performance)
1 (Default Mode)

Table 8: Performance of the pipeline decoding video at 1920x1080@30

Operation Mode	CPU (%)	GPU 1 (%)	GPU 2 (%)	FPS	latency (ms)
0 (max performance)
1 (min power)