Jetson Nano
Jetson Nano是NVIDIA推出的小型嵌入式AI计算开发板，面向边缘计算、机器人、智能相机等应用场景。该设备搭载NVIDIA GPU架构，支持并行AI推理计算，能够在低功耗条件下运行深度学习框架（如TensorFlow、PyTorch）及计算机视觉任务。其紧凑的外形与相对低廉的成本使其广泛用于研究、教育及嵌入式AI原型开发领域。
Core Facts
Timeline (last 90 days)
Edge computing modules such as the Raspberry Pi and Jetson Nano can be deployed directly alongside production lines to perform real-time inference.
The Jetson Nano has a compute performance of 472 GFLOPS, and the Jetson AGX Orin has a compute performance of 275 TOPS INT8.
The Jetson Nano was released in 2019
The older Jetson Nano supports only up to JetPack 4.6.x, corresponding to CUDA 10.2, which prevents newer framework versions such as PyTorch 2.x and TensorFlow 2.15+ from being natively compiled and run
For projects that process high-resolution video streams, run larger models (YOLOv8m/l), or require real-time detection above 30 FPS, the Jetson Nano will become a significant bottleneck.
An unoptimized PyTorch model may achieve only 2–3 FPS on the Jetson Nano, which can be improved to 10–25 FPS after TensorRT optimization, an improvement of up to 5–8x
The Jetson Nano running lightweight models such as YOLOv5s or YOLOv8n typically achieves 10–25 FPS after TensorRT optimization
The NVIDIA Jetson Nano has a compute performance of 472 GFLOPS and a power consumption of 5–10 W; the Jetson AGX Orin has a compute performance of 275 TOPS and a power consumption of 15–60 W.
All Facts (8)
The NVIDIA Jetson Nano has a compute performance of 472 GFLOPS and a power consumption of 5–10 W; the Jetson AGX Orin has a compute performance of 275 TOPS and a power consumption of 15–60 W.
65%UnverifiedEdge computing modules such as the Raspberry Pi and Jetson Nano can be deployed directly alongside production lines to perform real-time inference.
50%UnverifiedThe Jetson Nano has a compute performance of 472 GFLOPS, and the Jetson AGX Orin has a compute performance of 275 TOPS INT8.
50%UnverifiedThe Jetson Nano was released in 2019
50%UnverifiedThe older Jetson Nano supports only up to JetPack 4.6.x, corresponding to CUDA 10.2, which prevents newer framework versions such as PyTorch 2.x and TensorFlow 2.15+ from being natively compiled and run
50%UnverifiedAn unoptimized PyTorch model may achieve only 2–3 FPS on the Jetson Nano, which can be improved to 10–25 FPS after TensorRT optimization, an improvement of up to 5–8x
50%UnverifiedFor projects that process high-resolution video streams, run larger models (YOLOv8m/l), or require real-time detection above 30 FPS, the Jetson Nano will become a significant bottleneck.
50%UnverifiedThe Jetson Nano running lightweight models such as YOLOv5s or YOLOv8n typically achieves 10–25 FPS after TensorRT optimization
50%