What is edge computing?


We may be living in an artificial intelligence (AI) software bubble, but bringing AI outside the data center is a major challenge that remains to be solved. Edge AI computing is one manifestation of AI at the hardware level, where embedded systems at the network edge execute AI algorithms as part of a larger network. These embedded computing systems are getting smaller and more powerful thanks to improvements in embedded software, hardware platforms, and the ease of designing these systems.

Over the past decade, computing has shifted from desktops and mainframes to the cloud, and now computing power is moving from the cloud to the network edge. Just as computing power is being embedded in systems at the edge, AI tasks that used to be confined to the data center can now be performed in IoT devices and embedded systems at the network edge. These systems can collect data, process data, send data back to the cloud, and execute decisions as part of a larger AI-driven architecture.

Edge AI computing brings learning and inferencing outside of the data center.


Every computer network consists of nodes and endpoints, including endpoints at the edge of the network. AI at the edge involves implementing and executing AI algorithms at the network edge. These systems can be small IoT devices or larger embedded computers, and they may be involved in acquisition and pre-processing of data. This may occur without an active connection to the cloud or a larger network, or these systems may send their data back to the cloud as part of a larger AI ecosystem.


An edge device can be a large or small computing platform, including an IoT device. These devices can take a variety of form factors and have an array of functionality. AI edge devices must have high processing power, network connectivity, and sensor interfaces for data collection. IoT devices for AI at the edge can be implemented on a number of proven hardware platforms:

Single-board computers


Development boards

IoT Gateways

The MegaDrive board from Gumstix supports the Jetson Nano/Xavier NX modules for AI at the edge applications.

Because AI computing workloads are offloaded to the device level, many AI edge devices may be run as standalone IoT modules with intermittent network connectivity. With the right hardware and software stack, these modules can run AI algorithms autonomously, enabling multiple use cases that formerly required constant high-speed network connections.

What is AI at the Edge?


TensorFlow is an open source software library originally developed by Google. This framework gives programmers an easy way to build machine learning models, as well as integrate them into a larger AI algorithm. TensorFlow can be brought into a larger edge AI computing application written in Python, C++, and CUDA, which can then be deployed on embedded Linux kernels. If you want to start innovating a new system for AI at the edge, you’ll likely use TensorFlow to develop and implement machine learning models within your software stack.

Tensorflow Setup Demo


Deploying AI models on edge computing systems enables a range of applications based on the two fundamental inference tasks in machine learning: classification and prediction.

Some examples include:

Environmental monitoring

Smart home products

Sensor fusion

Object recognition from images or video


Models built on large datasets may need to be trained in the cloud, depending on the hardware capabilities of your endpoints.

New hardware platforms are enabling training of machine learning models at the edge.

Edge servers can be used to aggregate data from multiple endpoints and train models, which can then be deployed to edge AI devices in any of the above applications.

The CM4 PoE Smart Camera from Gumstix , With quad-core Raspberry Pi CM4, Coral Tensor processing unit, and power and Gigabit over Ethernet, is ready for AI deployment.

Designing embedded systems for edge computing and AI is now easier than ever thanks to modular design tools. With these powerful design tools, you can create a new IoT hardware platform for edge AI computing using proven computing modules and development boards. You don’t need to be a PCB design expert to create a new IoT platform for AI at the edge.

If you’re ready to start designing your embedded systems for edge AI computing, the modular design tools in Upverter let you leverage a broad range of industry standard COMs and popular modules to create carrier boards for intelligent IoT products. You can then deploy a trimmed-down Linux kernel like Yocto to optimize your computing performance for AI workloads. You can also add wireless connectivity modules, an array of sensors, high resolution cameras, and much more.

Take a look at some Gumstix customer success stories or contact us today to learn more about our products, design tools, and services.