Performance benchmarking for NVIDIA-accelerated Isaac ROS packages.
Isaac ROS Benchmark builds upon the ros2_benchmark to provide configurations to benchmark Isaac ROS graphs. Performance results that measure Isaac ROS for throughput, latency, and utilization enable robotics developers to make informed decisions when designing real-time robotics applications. The Isaac ROS performance results can be independently verified, as the method, configuration, and data input used for benchmarking are provided.
The ros2_benchmark playback node plug-in, for type adaptation and
negotiation, is provided for
NITROS, which
optimizes the performance of message transport costs through
RCL with GPU accelerated graphs of
nodes.
The datasets for benchmarking are explicitly not downloaded by default. To pull down the standardized benchmark datasets, refer to the ros2_benchmark Dataset section.
Please visit the Isaac ROS Documentation to learn how to use this repository.
Update 2025-10-24: Added new benchmarks for Isaac ROS 4.0
| Node |
Input Size |
AGX Thor |
x86_64 w/ RTX 5090 |
|---|---|---|---|
| AprilTag Node |
720p |
392 fps 10 ms @ 30Hz |
596 fps 2.3 ms @ 30Hz |
| FoundationPose Pose Estimation Node |
720p |
3.92 fps 260 ms @ 30Hz |
10.1 fps 89 ms @ 30Hz |
| DNN Stereo Disparity Node Full |
576p |
178 fps 22 ms @ 30Hz |
350 fps 5.6 ms @ 30Hz |
| DNN Stereo Disparity Node Light |
288p |
350 fps 9.4 ms @ 30Hz |
350 fps 5.0 ms @ 30Hz |
| Stereo Disparity Node |
1080p |
217 fps 51 ms @ 30Hz |
863 fps 4.1 ms @ 30Hz |
| Rectify Node |
1080p |
1800 fps 3.4 ms @ 30Hz |
2500 fps 2.8 ms @ 30Hz |
| TensorRT Node DOPE |
VGA |
180 fps 7.1 ms @ 30Hz |
321 fps 5.2 ms @ 30Hz |
| Triton Node DOPE |
VGA |
164 fps 32 ms @ 30Hz |
281 fps 4.5 ms @ 30Hz |
| TensorRT Node PeopleSemSegNet |
544p |
589 fps 7.6 ms @ 30Hz |
557 fps 3.1 ms @ 30Hz |
| Triton Node PeopleSemSegNet |
544p |
365 fps 9.6 ms @ 30Hz |
475 fps 5.1 ms @ 30Hz |
| DNN Image Encoder Node |
VGA |
416 fps 13 ms @ 30Hz |
446 fps 9.4 ms @ 30Hz |
| Occupancy Grid Localizer Node |
~50 sq. m |
28.3 fps 58 ms @ 30Hz |
50.1 fps 12 ms @ 30Hz |
| H.264 Decoder Node |
1080p |
596 fps 10 ms @ 30Hz |
596 fps 4.0 ms @ 30Hz |
| H.264 Encoder Node I-frame Support |
1080p |
390 fps 11 ms @ 30Hz |
596 fps 3.9 ms @ 30Hz |
| H.264 Encoder Node P-frame Support |
1080p |
429 fps 8.6 ms @ 30Hz |
596 fps 4.4 ms @ 30Hz |
| Nvblox Node |
– |
4.94 fps 89.8 ms |
4.96 fps 23.1 ms |
| Graph |
Input Size |
AGX Thor |
x86_64 w/ RTX 5090 |
|---|---|---|---|
| AprilTag Graph |
720p |
366 fps 13 ms @ 30Hz |
596 fps 4.1 ms @ 30Hz |
| Centerpose Pose Estimation Graph |
VGA |
50.5 fps 50 ms @ 30Hz |
50.2 fps 16 ms @ 30Hz |
| DOPE Pose Estimation Graph |
VGA |
138 fps 24 ms @ 30Hz |
199 fps 14 ms @ 30Hz |
| DNN Stereo Disparity Graph Full |
576p |
73.6 fps 29 ms @ 30Hz |
348 fps 8.5 ms @ 30Hz |
| DNN Stereo Disparity Graph Light |
288p |
219 fps 17 ms @ 30Hz |
350 fps 7.3 ms @ 30Hz |
| Stereo Disparity Graph |
1080p |
223 fps 5.6 ms @ 30Hz |
634 fps 6.6 ms @ 30Hz |
| DetectNet Object Detection Graph |
544p |
143 fps 27 ms @ 30Hz |
242 fps 17 ms @ 30Hz |
| Grounding DINO Object Detection Graph |
544p |
22.9 fps 70 ms @ 30Hz |
144 fps 15 ms @ 30Hz |
| RT-DETR Object Detection Graph SyntheticaDETR |
720p |
219 fps 23 ms @ 30Hz |
457 fps 8.0 ms @ 30Hz |
| TensorRT Graph PeopleSemSegNet |
544p |
460 fps 15 ms @ 30Hz |
510 fps 12 ms @ 30Hz |
| SAM Image Segmentation Graph Full SAM |
720p |
2.23 fps 270 ms @ 30Hz |
20.8 fps 58 ms @ 30Hz |
| SAM Image Segmentation Graph Mobile SAM |
720p |
21.0 fps 580 ms @ 30Hz |
74.1 fps 26 ms @ 30Hz |