On stage today at Google Cloud Next, Antony Passemard, Head of Product Management, IoT, Google Cloud will debut Edge TPU and Cloud IoT Edge:
Edge TPU is Google’s purpose-built ASIC chip designed to run TensorFlow Lite ML models at the edge. Google tells us these are high throughput systems built for something as demanding as inferring things from streaming video at your home (No, its not being used on Nest) to determine if actions are needed. While there is a SD card on the device so that you can store the data, we are told that this isn’t meant for training your ML models as much as making inferences based on them.
Cloud IoT Edge seems to be Google’s data processing machine learning heavy competitor to Amazon Web Services Lambda at edge, where lambda at edge is built to make decisions on the Cloudfront level with very quick speed, Cloud Iot Edge seems to be built to influence gateways, cameras, and end devices, this service probably built first due to Google’s work inside the device space being much larger.
By running on-device machine learning models, Cloud IoT Edge with Edge TPU provides a internet free inference, which as any computer scientist will tell you means it will be significantly faster than general-purpose IoT gateways. Even better is by not sending the data out over the internet you don’t have to give state actors access to a live video stream of your home, which is nice.
The Edge TPU development kit:
If you want one of these in early access, no need to dig further, sign up via this form.
This kit includes a system on module (SOM) that combines Google’s Edge TPU, a NXP CPU, Wi-Fi, and Microchip’s secure element in a compact form factor. It’ll will be available to developers this October. Do of course let us know what you all are able to build with this, as we all love seeing new hardware!
The Edge TPU development kit — SOM (above) and base board (below) (photo google)