Today’s video is all about the new member of the Nvidia Jetson family, the Jetson Orin Nano Developer Kit. This developer kit is considered to be the most powerful of all the Jetson family options and comes with a variety of features and capabilities. The kit, which includes a power supply, is equipped with various ports such as DisplayPort, USB, Ethernet, and more.
The Jetson Orin Nano Developer Kit includes the Jetson module, a carrier board with all the useful ports, and a heatsink, making it a sophisticated adapter for the Jetson. While the kit itself is powerful, an added bonus is that it can be upgraded with a more powerful Jetson module for even greater performance.
In terms of performance, the Jetson Orin Nano Developer Kit offers impressive capabilities, including the use of the latest GPU architecture, advanced CPU, and the ability to run machine learning models and benchmarks with high speed and efficiency.
The video also includes demonstrations of different models and their capabilities, showcasing the kit’s ability to detect people, improve semantic segmentation, and retrain models for specific tasks such as robotics and security systems.
Overall, the Jetson Orin Nano Developer Kit offers outstanding performance and versatility, making it an ideal choice for a wide range of AI and robotics applications. With its upgradeable design and powerful features, it is a worthwhile investment for developers and enthusiasts looking to push the boundaries of AI and robotics technology.
Watch the video by Skyentific
Video Transcript
Hello and welcome to my Channel today I would like to talk about new member of Nvidia Judson family Judson Oren Nana developer kit here there are some parameters but we’re going to talk about them later the Box includes power supply here for example I have some other Jetsons like Jetson Nana two gigabyte
This is original jetsonana 4 gigabyte reaction this is Jetson Orin ajx Orem developer kit this is the most powerful Jetson which you can buy but also the most expensive one and this one is way more powerful than this one or this one but less expensive than this one so let’s unbox it tada
And here’s the power supply I would like to clarify a small scene that the Jetson itself is actually this small board over here which is installed on the bigger carrier board and this carrier board it has all the ports all the useful ports so basically the carrier board is just a
Sophisticated adapter for the Jetson the carrier board plus Jetson plus heatsink is called developer kit and here you can replace this Jetson or in Nana for the more powerful Jetson or in NX so there is a possibility for kind of upgrade but this kind of strange upgrade because you
Keep the cheap part and you replace the expensive part but it’s possible this is for power supply DisplayPort 4 USB 3.2 Gen 2 gigabit Ethernet USBC for the debug and device mode to CSI 22 pin camera connectors connector for the optional buttons like reset and power off and place for the SD card
40 pin gpio connector and on the other side there are m.2 key pcie connector which is populated with the Wi-Fi module and here’s ratu antennas m.2 km falling PC gen 3 and m.2 km two lane PCA gen 3 and here there is a schematics with a all the ports
So the power is provided by 19 volts power supply so actually this camera ports are similar to the ports from the Raspberry Pi zero so this means that there are a lot of ribbon cables which are compatible with this port and with the standard Raspberry Pi compatible
Camera let’s connect it to the keyboard to the screen and to the power supply put the SD card I already prepared the SD card over there and let’s see how it runs so let’s switch it on for the first time the LED lights up the first boot is standard for the Ubuntu
And I use default values everywhere by default it runs the power mode at 15 watts another option is 7 Watt so here there is a less power and also less performance here the max performance and max power so here are the Jetson developer kits which are offered by Nvidia so at the
Lower level of this offering there is Jetson Nana developer kit the original one it has the previous architecture of the GPU at the high and there is just an egx orange the super powerful quite expensive but a really good device and in intermediate position there is Jason
Or a Nana developer kit and here you can see the performance on all of them the great thing about them that they use the same software jetpack and that’s why you can start with the Jetson Nana with the original one and if you need the power you can go to the more expensive
Versions and here the Jetson modules so not the developer kits but just modules so these are the modules and for the Jetson audience so with the latest architecture of the GPU so there is Jetson orenana 8 gigabyte and 4 gigabyte version on the developer it’s 8 gigabyte
Version there is Jetson or in NX more powerful one and Jetson agx Orange it’s a super powerful one and so you can see that the cheapest one they don’t require a lot of power and the most powerful One require up to 60 watts which is still quite lower for the robotics application
So it’s a perfectly fine with 60 watts on this developer kit which we have we can replace the original Jetson Oren Nana module with this one Jetson or in NX module but in this case you would need to pay 400 more for the 8 gigabyte version and
600 more for the 16 gigabyte version but at the same time you have more than twice gain on the performance compared to the Judson Orlando and here’s the other specs for this Jetson orenana the main specs it use the latest architecture of the GPU 1024 Cuda cores up to 80 times
Performance of the original Jetson Nana 6 core arm CPU and it’s going to be available soon unfortunately on the Jetson Oren Nana series there is no video encoder Hardware video encoder so you have to use the CPU for the encoding of the video but still this
Works so there is no major problem with this for the storage you can use the SD card as original Jetson Nana but also you can use the external nvme storage I think this was the most important specs and now let’s try some demos which was prepared by Nvidia first of all let’s
Run some benchmarks and here’s a different models and the speed with which it can run these models so the frames per second how quickly it can process these different models so these values are quite impressive in the next demo we’re going to run the people.net model and this model is
Transformer model so it’s a model which uses the same kind of Technology as all the super popular uh AI models now like charge Deputy Etc and so here as you can see this model can detect people from the camera and actually it’s quite impressive that it can detect people
From the camera which looks uh at some weird angle and all the people are a little bit tilted this is not obvious eight frames per second it’s not super fast but for the security system or something like this it should be enough the next demo is super interesting and
The main idea is that we take a robot which is super small super low and it has a camera on it so it looks from the bottom on the top and so like this it sees the people at the kind of weird angle and most of the standard models
Are not trained at this weird angle and that’s why the semantic segmentation is not working perfectly just out of the box and so as you can see over here how this model works out of the box you see that it works but there is a room for improvements and afterwards this model
Was retrained on the synthetic data and this synthetic data comes from the applicator from the Isaac seam and so with this additional data it was retrained using the Tau and after this model was deployed on the Jetson Orient and here you can see the result
Of this model and as you can see it works uh way better and I should say that this process of generating the new synthetic data and the retraining the model it was actually quite a quick computational time was a little bit more than one hour and as you can see like
This we can improve this original model quite significantly and it works better I think this is a really cool example which shows first of all how you can use the Isaac Sim replicator how you can use the Tau and how you can quite quickly improve the model for your specific task
I really love this demo I have found this ribbon cable so this is a ribbon cable for the Raspberry Pi zero so this part goes to the camera and this part goes to your Edge device I also have a wide view camera from my old project I know that this camera works perfectly
With original Jetson Nana so one side of the ribbon cable goes here and the second part goes to the Jetson or in Nana the camera is installed so for this I need to run this command and now theoretically the camera is recording stop recording let’s see the result it’s
Upside down but it doesn’t matter it looks like it works I have connected the Raspberry Pi camera this is a camera version 2.1 Let’s test it quickly it’s not a great camera but it should work let’s see the file I was moving it a little bit too fast
But it was working in my opinion this Jetson Oren Nana developer kit is really useful for the DIY Community it’s small it doesn’t require a lot of power and so that’s why it’s perfect for the mobile robotics or for the smaller robots or for the robots where you don’t want to
Have a bulky desktop PC I already decided to use this module for some speech recognition so most probably you will see the video about this in the near future thank you for watching this video till the end huge thank you to people who support me via patreon and Via YouTube channel membership
Here’s their names thank you as usual stay safe good luck with your projects and see you next time what is NVIDIA Jetson and why I should use it Nvidia Jetson is a series of high performance embedded Computing systems designed for Edge processing and AI applications these devices are compact
Energy efficient and feature powerful Nvidia gpus and multi-core arm CPUs making them ideal for running complex AI algorithms and processing large amounts of data in real time there are several reasons why you should use Nvidia Jetson for your AI projects one high performance Jetson devices are powered by Advanced gpus that can
Deliver high performance Computing required for running AI models 2. low power consumption Jetson devices are designed to consume low power making them ideal for use in energy efficient Edge devices three cost effective Jetson devices offer high performance Computing capabilities at a much lower price compared to traditional workstations 4.
Easy to use Jetson devices come with a user-friendly development environment and software tools that enable quick and easy development of AI applications 5. compatibility Jetson devices are compatible with commonly used AI Frameworks such as tensorflow Pi torch and Cafe enabling developers to take advantage of a wide range of pre-trained models and algorithms
Video “Compact AI brain for your robot (Jetson Orin Nano Developer Kit)” was uploaded on 04/14/2023 to Youtube Channel Skyentific