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	AGX Xavier	Orin NX	Orin Nano 8GB	x86_64 w/ RTX 4060 Ti
Visual SLAM Node	720p	228 fps 40 ms	122 fps 58 ms	127 fps 74 ms	113 fps 65 ms	456 fps 37 ms
AprilTag Node	720p	216 fps 9.7 ms	167 fps 14 ms	106 fps 14 ms	74.1 fps 21 ms	473 fps 6.4 ms
Freespace Segmentation Node	576p	1760 fps 1.2 ms	980 fps 3.8 ms	1410 fps 1.6 ms	1060 fps 2.3 ms	3500 fps 0.32 ms
Depth Segmentation Node	576p	47.7 fps 43 ms	25.3 fps 120 ms	30.0 fps 98 ms	–	89.9 fps 28 ms
DNN Stereo Disparity Node Full	576p	78.8 fps 4.1 ms	26.2 fps 15 ms	27.2 fps 6.2 ms	22.1 fps 7.4 ms	204 fps 4.4 ms
DNN Stereo Disparity Node Light	288p	288 fps 5.6 ms	109 fps 7.3 ms	128 fps 5.6 ms	86.7 fps 6.9 ms	350 fps 4.2 ms
Stereo Disparity Node	1080p	160 fps 11 ms	88.6 fps 18 ms	76.5 fps 19 ms	52.4 fps 26 ms	357 fps 7.2 ms
Rectify Node	1080p	819 fps 5.4 ms	569 fps 5.9 ms	496 fps 5.3 ms	347 fps 6.4 ms	900 fps 3.9 ms
TensorRT Node DOPE	VGA	48.2 fps 22 ms	13.4 fps 80 ms	18.5 fps 56 ms	13.0 fps 81 ms	103 fps 11 ms
Triton Node DOPE	VGA	47.8 fps 23 ms	14.5 fps 830 ms	20.4 fps 540 ms	14.5 fps 790 ms	99.0 fps 10 ms
TensorRT Node PeopleSemSegNet	544p	600 fps 3.4 ms	342 fps 12 ms	324 fps 5.6 ms	215 fps 6.4 ms	721 fps 2.2 ms
Triton Node PeopleSemSegNet	544p	311 fps 4.0 ms	177 fps 12 ms	194 fps 5.7 ms	–	682 fps 2.0 ms
DNN Image Encoder Node	VGA	1120 fps 1.6 ms	547 fps 3.6 ms	1070 fps 2.3 ms	–	3020 fps 0.66 ms
Occupancy Grid Localizer Node	~50 sq. m	15.9 fps 64 ms	9.45 fps 290 ms	7.34 fps 140 ms	5.17 fps 200 ms	50.1 fps 20 ms
H.264 Decoder Node	1080p	179 fps 9.5 ms	151 fps 12 ms	–	–	596 fps 3.1 ms
H.264 Encoder Node I-frame Support	1080p	402 fps 12 ms	164 fps 19 ms	–	–	596 fps 4.8 ms
H.264 Encoder Node P-frame Support	1080p	464 fps 11 ms	397 fps 13 ms	–	–	596 fps 4.1 ms

Graph	Input Size	AGX Orin	AGX Xavier	Orin NX	Orin Nano 8GB	x86_64 w/ RTX 4060 Ti
AprilTag Graph	720p	213 fps 14 ms	122 fps 21 ms	102 fps 20 ms	71.0 fps 26 ms	472 fps 9.5 ms
Freespace Segmentation Graph	576p	45.9 fps 41 ms	19.2 fps 130 ms	27.6 fps 95 ms	21.3 fps 110 ms	91.0 fps 30 ms
Centerpose Pose Estimation Graph	VGA	36.1 fps 5.7 ms	20.2 fps 16 ms	19.4 fps 7.4 ms	13.8 fps 12 ms	50.2 fps 14 ms
DOPE Pose Estimation Graph	VGA	39.8 fps 33 ms	12.5 fps 160 ms	17.3 fps 120 ms	–	89.2 fps 15 ms
DNN Stereo Disparity Graph Full	576p	74.0 fps 20 ms	21.0 fps 93 ms	26.1 fps 42 ms	–	191 fps 11 ms
DNN Stereo Disparity Graph Light	288p	260 fps 13 ms	101 fps 20 ms	116 fps 16 ms	–	350 fps 12 ms
Stereo Disparity Graph	1080p	155 fps 15 ms	84.7 fps 21 ms	74.7 fps 21 ms	49.0 fps 31 ms	357 fps 3.0 ms
DetectNet Object Detection Graph	544p	232 fps 11 ms	90.0 fps 19 ms	105 fps 15 ms	74.2 fps 22 ms	644 fps 5.6 ms
TensorRT Graph PeopleSemSegNet	544p	421 fps 8.1 ms	252 fps 12 ms	238 fps 9.6 ms	162 fps 13 ms	704 fps 5.5 ms

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.