Matthijs van der Veer's Blog

IoT Edge AI: Top Single-Board Computers for Custom Vision

March 22, 2023

So, you’re looking to run Custom Vision on IoT Edge and need a single-board computer (SBC) that can keep up with the workload? You’re in the right place! In this blog post, we’ll delve into some great SBC options that can deliver low-latency performance for your AI at the edge projects. Let’s jump in and explore NVIDIA Jetson devices, Google Coral, BeagleBone, and ODROID as potential solutions.

Why the Raspberry Pi Just Doesn’t Cut It

Now, the Raspberry Pi is a fantastic little device, but it has its limitations. If you’ve ever tried running a Custom Vision model on it, you’ll know what I mean – slow response times! And it’s all because the Raspberry Pi lacks a dedicated GPU, which is an absolute must for handling the intense computations involved in machine learning tasks.

Meet the NVIDIA Jetson Family and Other SBC Alternatives

So, what’s the solution? Say hello to the NVIDIA Jetson family and other powerful SBC options! These devices are specifically designed for AI and machine learning applications, making them a perfect choice for anyone looking to improve latency and frame rate.

Jetson Nano

Starting with the Jetson Nano, this nifty device is an ideal step up from the Raspberry Pi. Packing a 128-core NVIDIA Maxwell GPU, the Nano can deliver up to 472 GFLOPs of performance. What does that mean for you? Running Custom Vision at IoT Edge with latencies and frame rates closer to your target of 2-3 FPS. Not bad, right?

Jetson Xavier NX

But what if you need even more power? Enter the Jetson Xavier NX. This beast features a 384-core NVIDIA Volta GPU with 48 Tensor Cores, offering a whopping 21 TOPS of AI performance. If you’re looking for faster results for Custom Vision at IoT Edge, this might just be the perfect choice.

Google Coral Dev Board

The Google Coral Dev Board is an SBC that features the Edge TPU, a small ASIC designed by Google for high-performance machine learning inference. With its 4GB LPDDR4 RAM and NXP i.MX 8M SOC, this board can run TensorFlow Lite models efficiently and is a suitable option for edge AI applications.

BeagleBone AI

The BeagleBone AI is an SBC designed for AI applications at the edge. It comes equipped with a Texas Instruments AM5729 SoC, which includes dual ARM Cortex-A15 CPU cores, dual C66x DSP cores, and four embedded-vision-engine (EVE) cores for hardware-accelerated machine learning. This makes it a viable option for running Custom Vision at IoT Edge with improved performance.

ODROID-XU4

The ODROID-XU4 is an SBC powered by the Samsung Exynos 5422 SoC, which features an ARM Mali-T628 MP6 GPU. This board offers better GPU performance compared to the Raspberry Pi and might be capable of running Custom Vision at IoT Edge with reduced latency.

Conclusion

When it comes to finding a single-board computer for running Custom Vision at IoT Edge, it’s crucial to pick a device with a dedicated GPU. With options like the NVIDIA Jetson family, Google Coral Dev Board, BeagleBone AI, and ODROID-XU4, you’re sure to find a device that meets your performance needs. So, go ahead and give your Custom Vision experience at the edge a major boost with the right device!


Matthijs van der Veer

Written by Matthijs van der Veer, Coding Architect @ Xebia & Microsoft MVP Award Winner. He infrequently tweets on Twitter, occasionaly pushes code to GitHub and often answers IoT related questions on StackOverflow. Feel free to connect on LinkedIn. Also Mastodon is a thing now.