Introductory tutorials, case source code, learning materials, readership
Please visit: python666.cn
Hello everyone, I am Crossin. Welcome to Crossin's programming classroom!
Today I bring you an unboxing review video.
Everyone knows that AI is very popular recently. Even if you don’t care about the news in the technology circle, you must have seen content such as ChatGPT and AI painting more than once.
And what I want to experience today is an AI-related hardware device.
Without further ado, let's take a look.
Today's protagonist is the latest artificial intelligence computer released by Nvidia at the GTC conference:
NVIDIA Jetson Orin Nano Developer Kit
This developer kit mainly consists of 3 parts, an Orin Nano 8G module, a carrier board and a fan.
The main interfaces on the carrier board include the power interface, the DP interface connected to the display device, 4 USB interfaces, the network cable interface, and the type-c interface.
There are also 40 pins on both sides for connecting some sensors and peripherals, as well as the interface for connecting high-speed cameras.
There is also a micro-SD card slot on the back for external storage.
In one of my previous videos, I used the NVIDIA Jetson AGX Orin machine in an action recognition project , which is also a product suitable for AI edge computing. It can be seen intuitively that this time the Orin Nano is much smaller in size and lighter in weight. So it will be more suitable for some application scenarios that require high space and portability, such as small robots and drones.
So what about its performance? Let's experience it on the real machine and run a score to try it out.
System and environment installation via Micro-SD card pre-written with NVIDIA JetPack SDK. After the installation is complete, we download the benchmark test file from the official website, and after running the script to install the environment, we can run the benchmark test.
This is a test related to the visual AI model. After running for about 45 minutes, the results are as follows.
In the previous AGX orin video, we also ran it once. Compared with the previous results, the performance of AGX orin is roughly 3~5 times that of orin nano. But such a comparison is obviously unfair, regardless of size or price, including power input power, AGX is several times larger. The demand scenarios for the two are not the same.
But if we compare it with the old Jetson nano, we will find that the performance of Orin Nano is an order of magnitude improvement. According to official data, Orin Nano has achieved an AI performance of 40 trillion operations per second (TOPS), which is 80 times that of the previous nano.
Let's look at a more intuitive demo. A people detection program based on PeopleNet Transformer. The official docker container is provided for this demo, so only a few lines of commands are needed to complete the download, configuration and operation.
The running frame rate is about 7 to 8 frames per second. Although this level cannot reach real-time, it is enough for many application scenarios. If you have higher requirements, you can consider AGX orin, and the efficiency of the same demo can reach 30 frames per second.
It can be seen from the above introduction and test that the Jetson Orin Nano
Small size, light weight, low power consumption , more suitable for some harsh edge computing scenarios
But small and small, its performance is not weak , and it can provide AI performance of 40 trillion operations , which is 80 times that of the previous JetsonNano. Competent for a large number of AI-related development needs.
In addition, it is also equipped with NVIDIA JetPack 5.1 , which provides related libraries for accelerating GPU computing, multimedia, computer vision and graphics, supports SDKs such as DeepStream, Isaac, Riva, and is a complete edge AI development environment. It greatly facilitates developers.
So NVIDIA Jetson Orin Nano is very suitable for developing entry-level AI applications. In addition, it is also a good choice to use it as a device for learning and practicing AI. But if you have higher requirements for performance, you may have to consider AGX orin.
In addition to the programs just demonstrated, there are many interesting AI demos on Jetson, which can be easily built and run, as well as develop your own AI applications on this basis. I won't demonstrate them one by one here.
If you have any thoughts on the development of AI, welcome to discuss in the message area.
Thank you for retweeting and liking ~
_Previous article recommendation_
It took 2 days, I made a somatosensory game console
If you need to know about paid quality courses and teaching Q&A services
Please reply in Crossin's programming classroom : 666
