The last few years have held some exciting announcements from NVIDIA. Their Jetson line is widely considered one of the most powerful edge AI solutions available in the market today. Their latest iteration, the Orin family of products, has allowed them to do even more with fewer watts and in smaller form factors. Needless to say, there were a lot of folks waiting to get their hands on these products to test them for themselves with balena.
If you’re not familiar with NVIDIA’s Jetson edge AI family, you should take a look at my colleague Alan’s overview of the series, including which devices we support and best practices for using them with balenaCloud: https://blog.balena.io/how-to-use-nvidia-Jetson-devices-on-balena/
We’ve thankfully supported NVIDIA’s own Orin Development Kits for customers to go through their initial stages of testing for some time. Now we want to offer a solution that will help customers go as quickly as possible from “idea” to “production”, leading us to support a commercial off-the-shelf (COTS) Jetson Orin NX device. COTS products let you get to market quickly, giving you something that meets your needs and is production ready, while you work internally on how to optimize and scale over time.
While there are a number of models out in the market, we reached out to our Partners at Seeed Studio, who supplied us with a reComputer J4012 to test with. It’s got an NVIDIA Jetson Orin NX (up to 100 TOPS AI performance), and configured with 16GB RAM and an 128GB NVMe SSD, making it a good configuration to test some generative AI models you might have been curious about trying, among other things.
You can of course read NVIDIA’s full specifications for the Orin NX module, but let me share some of the really nice quality-of-life features that Seeed has added to the reComputer J4012 that might not be easily noticed by pictures and datasheets alone:
- The top access panel is magnetically attached – the magnets are strong enough to not let it pop off when jostled or being moved, but make life so much easier for those who are testing different configurations of hardware or servicing the product. (Fewer screwdriver tips to keep with you!)
- The bottom panel allows you to mount the J4012 using an 8.5 cm mounting pattern, so it can attach to a pole or hang on a wall just as easily as it can sit on a desk.
- There are also two 2-lane MIPI-CSI camera connectors for stereoscopic vision applications, which is a really nice option.
If you’ve ever used an NVIDIA Jetson product, you know there are quite a few steps involved in getting it provisioned the first time. Thankfully, my colleague Alan was kind enough to create a video that walks you step-by-step through the process. He gives a nice unboxing overview in the first two minutes and from there goes straight to adding a device to balenaCloud and getting his J4012 running with a PyTorch example application. Alan has a nice and soothing voice, so please enjoy his video!
Using the J4012 with balena
If you’re looking for a simple starter project to get familiar with the Seeed J4012 on balena, check out our PyTorch/TensorRT example project. This project is based on NVIDIA’s L4T PyTorch base image and includes a few example projects with detailed instructions in the repository for running them. There’s even an example of using Yolov3 and TensorRT to generate images containing object inferences with bounding boxes, as seen below:
Clik here to view.
You can deploy this project by clicking the button below. For information on how to flash the OS to the Seeed J4012, follow the instructions on the device deployment screen or see the video below!
Clik here to view.
These images are created especially for this device and include apt sources for installing the appropriate NVIDIA software tools. For more details about balena base images, click here.
If you are looking to create your own containers to run on this hardware, and want to build and install your own tools, check out our base images for the Seeed J4012.
What’s next?
Let us know in the comments if you have deployed a similar project or used any NVIDIA Jetson device with balena. If you have any questions about using the balena platform get in touch!
The post Announcing balenaCloud support for the Seeed reComputer J4012 with NVIDIA Jetson Orin NX module appeared first on balena Blog.