Google Coral M.2 Accelerator with Dual Edge TPU

The Coral M.2 Accelerator with Dual Edge TPU integrates two Edge TPUs into existing computer systems with the help of an M.2 E-key interface. This Edge TPU module is particularly suitable for mobile and embedded systems that can benefit from accelerated machine learning. Each Edge TPU coprocessor is capable of 4 trillion arithmetic operations per second (4 TOPS) with 2-watt power consumption. For example, modern Mobile Vision models such as MobileNet v2 can run efficiently at close to 400 FPS. With two Edge TPUs (and thus 8 TOPS) you can double the performance of the system - for example, by running two models in parallel, or by distributing the processing steps of a model between both Edge TPUs. The M.2 Accelerator with Dual Edge TPU is supported under Debian Linux based systems as well as under Windows 10. We also carry the Coral USB Accelerator - to be able to retrofit systems that do not offer an M.2 interface with a USB 3.0 interface. Specification TPU

* 2 x Google Edge TPU ML accelerator coprocessor

* 8 TOPS (int8); 2 TOPS per watt

Interface & software support

* Interface: M.2 E-key (with two PCIe Gen2 x1 lanes)

* Supports Linux, and Windows 10 on the host system

Other data

* Dimensions: 22 mm x 30 mm x 2.8 mm (M.2-2230-D3-E)

* Weight: 2.5 g

* Operating temperature: -40 to + 85 ° C

Requirements for the host system Linux:

* 64-bit Linux Debian 10.0 / Ubuntu 16.04 (or newer)

* CPU architecture: x86-64, or ARMv8 (64-bit)

or Windows:

* Windows 10 (64-bit),

* x86-64 CPU architecture

Google Part Number

* G650-06076-01

Important Notes Although the M.2 specification requires that E-key sockets provide two PCIe x1 lanes, most manufacturers only provide PCIe once! In order to be able to use both Edge TPUs, please make sure that your socket actually provides two PCIe x1. Please also note the information in the datasheet on peak current requirements (up to 3 A per EdgeTPU) and thermal management. Each Edge TPU contains a built-in temperature sensor and allows parameters to be configured when it should be switched off. Warning: Overheating of the system can lead to fire or the destruction of hardware! Potential for Industrial Applications The Coral Edge TPU is a revolutionary product for machine learning applications. This enables embedded solutions that can, for example, recognize problems with workpieces, recognize traffic situations and much more. Resources M.2 accelerator module

* Coral M.2 Accelerator with Dual Edge TPU data sheet (PDF)

* Getting started with the M.2 or PCIe Accelerator

* Temperature management of the PCIe module

Edge TPU & models

* Introduction to Models on the EdgeTPU

* Overview of inferencing on the Edge TPU

* Operation of multiple models with multiple Edge TPUs (English)

* Split a model across multiple Edge TPUs

API & Downloads

* PyCoral API (Python)

* Libcoral API (C ++)

* Libedgetpu API (C ++)

* Edge TPU compiler

* precompiled models

* All software downloads