Coral M.2 Accelerator with Dual Edge TPU

The Coral M.2 Accelerator with Dual Edge TPU brings powerful, on-device machine learning processing to compact and power-efficient applications. By incorporating two dedicated Google Edge TPU ML accelerators, this module delivers exceptional inference speeds for running complex TensorFlow Lite models directly on your hardware.

The key to its performance is the dual Edge TPU architecture. Each Edge TPU is a specialised ASIC designed by Google for optimal efficiency, capable of performing 4 trillion operations per second (4 TOPS) while consuming just 2 watts of power. This translates to an impressive 2 TOPS per watt. For perspective, a single Edge TPU can execute advanced mobile vision models like MobileNet v2 at nearly 400 frames per second.

This module features two independent PCIe Gen2 x1 interfaces, one for each TPU. This design enables you to double your inference throughput to a total of 8 TOPS. You can utilise this power in flexible ways, such as running two different models simultaneously in parallel or pipelining a single, more complex model across both accelerators for faster results.

By processing data locally on the device, the Coral M.2 Accelerator offers significant advantages. It drastically reduces latency compared to cloud-based processing, enhances data privacy by keeping sensitive information on-premises, and eliminates the dependency on a constant, high-bandwidth internet connection.

Housed in a compact M.2 E-key form factor (specifically an M.2-2230-D3-E module), it is designed for easy integration into a wide range of systems. With robust integrated power management and a wide operating temperature range, it is suitable for deployment in diverse environments from industrial edge devices to embedded AI solutions.