Performance Summary

The performance results of benchmarking the Isaac ROS packages on both Jetson and x86_64 platforms are below:

Node	Input Size	AGX Orin	Orin NX	Orin Nano 8GB	x86_64 w/ RTX 4090
AprilTag Node	720p	249 fps 4.5 ms @ 30Hz	116 fps 9.3 ms @ 30Hz	80.7 fps 14 ms @ 30Hz	596 fps 0.97 ms @ 30Hz
Freespace Segmentation Node	576p	2120 fps 1.7 ms @ 30Hz	2490 fps 1.6 ms @ 30Hz	1560 fps 2.3 ms @ 30Hz	3500 fps 0.52 ms @ 30Hz
Depth Segmentation Node	576p	45.8 fps 79 ms @ 30Hz	28.2 fps 99 ms @ 30Hz	–	105 fps 25 ms @ 30Hz
FoundationPose Pose Estimation Node	720p	1.72 fps 690 ms @ 30Hz	–	–	9.61 fps 110 ms @ 30Hz
DNN Stereo Disparity Node Full	576p	103 fps 12 ms @ 30Hz	42.1 fps 26 ms @ 30Hz	–	350 fps 2.3 ms @ 30Hz
DNN Stereo Disparity Node Light	288p	306 fps 5.6 ms @ 30Hz	143 fps 9.4 ms @ 30Hz	–	350 fps 1.6 ms @ 30Hz
Stereo Disparity Node	1080p	117 fps 11 ms @ 30Hz	78.4 fps 14 ms @ 30Hz	53.5 fps 21 ms @ 30Hz	978 fps 1.6 ms @ 30Hz
Rectify Node	1080p	1160 fps 2.2 ms @ 30Hz	595 fps 3.6 ms @ 30Hz	411 fps 4.4 ms @ 30Hz	2500 fps 0.70 ms @ 30Hz
TensorRT Node DOPE	VGA	47.9 fps 24 ms @ 30Hz	18.1 fps 56 ms @ 30Hz	13.1 fps 81 ms @ 30Hz	298 fps 4.6 ms @ 30Hz
Triton Node DOPE	VGA	47.2 fps 24 ms @ 30Hz	20.3 fps 530 ms @ 30Hz	14.5 fps 780 ms @ 30Hz	276 fps 4.6 ms @ 30Hz
TensorRT Node PeopleSemSegNet	544p	460 fps 4.1 ms @ 30Hz	348 fps 6.1 ms @ 30Hz	238 fps 7.0 ms @ 30Hz	–
Triton Node PeopleSemSegNet	544p	304 fps 4.8 ms @ 30Hz	206 fps 6.5 ms @ 30Hz	–	–
DNN Image Encoder Node	VGA	420 fps 12 ms @ 30Hz	382 fps 11 ms @ 30Hz	–	574 fps 5.2 ms @ 30Hz
Occupancy Grid Localizer Node	~50 sq. m	12.9 fps 86 ms @ 30Hz	8.36 fps 130 ms @ 30Hz	5.81 fps 190 ms @ 30Hz	50.1 fps 8.8 ms @ 30Hz
H.264 Decoder Node	1080p	197 fps 8.2 ms @ 30Hz	–	–	596 fps 4.2 ms @ 30Hz
H.264 Encoder Node I-frame Support	1080p	402 fps 13 ms @ 30Hz	–	–	409 fps 3.4 ms @ 30Hz
H.264 Encoder Node P-frame Support	1080p	473 fps 11 ms @ 30Hz	–	–	596 fps 2.1 ms @ 30Hz
Nvblox Node	–	4.87 fps 35.9 ms	4.95 fps -1.43 ms	4.95 fps 23.1 ms	4.95 fps 195 ms

Graph	Input Size	AGX Orin	Orin NX	Orin Nano 8GB	x86_64 w/ RTX 4090
AprilTag Graph	720p	246 fps 6.3 ms @ 30Hz	111 fps 12 ms @ 30Hz	77.5 fps 20 ms @ 30Hz	596 fps 1.6 ms @ 30Hz
Freespace Segmentation Graph	576p	36.5 fps 84 ms @ 30Hz	27.8 fps 100 ms @ 30Hz	22.0 fps 98 ms @ 30Hz	100 fps 22 ms @ 30Hz
Centerpose Pose Estimation Graph	VGA	47.1 fps 6.5 ms @ 30Hz	30.0 fps 50 ms @ 30Hz	19.9 fps 28 ms @ 30Hz	50.2 fps 12 ms @ 30Hz
DOPE Pose Estimation Graph	VGA	27.7 fps 16 ms @ 30Hz	17.8 fps 14 ms @ 30Hz	–	187 fps 7.8 ms @ 30Hz
DNN Stereo Disparity Graph Full	576p	33.5 fps 25 ms @ 30Hz	35.2 fps 34 ms @ 30Hz	–	350 fps 5.6 ms @ 30Hz
DNN Stereo Disparity Graph Light	288p	179 fps 14 ms @ 30Hz	126 fps 15 ms @ 30Hz	–	350 fps 4.4 ms @ 30Hz
Stereo Disparity Graph	1080p	111 fps 15 ms @ 30Hz	72.2 fps 19 ms @ 30Hz	49.7 fps 26 ms @ 30Hz	695 fps 3.9 ms @ 30Hz
DetectNet Object Detection Graph	544p	70.5 fps 26 ms @ 30Hz	30.1 fps 46 ms @ 30Hz	22.9 fps 57 ms @ 30Hz	254 fps 11 ms @ 30Hz
RT-DETR Object Detection Graph SyntheticaDETR	720p	56.5 fps 30 ms @ 30Hz	33.8 fps 39 ms @ 30Hz	24.1 fps 53 ms @ 30Hz	490 fps 7.1 ms @ 30Hz
TensorRT Graph PeopleSemSegNet	544p	371 fps 19 ms @ 30Hz	250 fps 20 ms @ 30Hz	163 fps 23 ms @ 30Hz	–
SAM Image Segmentation Graph Full SAM	720p	2.22 fps 390 ms @ 30Hz	–	–	16.4 fps 280 ms @ 30Hz
SAM Image Segmentation Graph Mobile SAM	720p	8.75 fps 570 ms @ 30Hz	5.34 fps 1400 ms @ 30Hz	2.22 fps 340 ms @ 30Hz	68.6 fps 23 ms @ 30Hz

Live Graph	Input Size	Nova Carter
Data Recorder Live Graph 4 Hawk Cameras	1200p	22.0 fps (per stream avg) 0 dropped frames (avg)
Multicam Visual SLAM Live Graph 4 Hawk Cameras	1200p	30.1 fps
DNN Stereo Disparity Live Graph 3 Hawk Cameras 1x Full ESS and 2x Throttled Light ESS	1200p	Full: 30.2 fps Light: 15.2 fps (avg)
Perceptor Graph 3 Hawk Cameras	1200p	Nvblox ESDF: 9.46 fps Nvblox Mesh: 1.01 fps Visual Odometry: 30.0 fps

Methodology

The performance numbers provided above are measured using Isaac ROS Benchmark. Configuration files for reproducing results are provided in the launch scripts for each benchmark here.

Frames per second (fps) are measured using the maximum sustained framerate for the accelerated computational pipeline on each listed platforms. Performance measured includes input node → graph of node(s) under performance test → output node where the publishing rate of the input node is auto-tuned to discover the peak throughput dropping <5% of the frames. Average fps computed over 5 runs, discarding minimum and maximum runs; latency measured at 30hz publishing rate matching sensor capture rate.