Google Coral M.2 gas pedal (A+E)
44,10 € INCL. VAT
- Google Edge TPU coprocessor for AI/ML acceleration.
- M.2 A+E Key interface via PCIe Gen2 x1.
- Supports TensorFlow Lite for running optimized ML models.
Entreprise, école ou collectivité ? business account Offres éducation
- Description
- Additional information
Description
Google Coral M2 A+E: the M.2 Edge TPU accelerator for your embedded systems
The Google Coral M.2 Accelerator with A+E key integrates Google's Edge TPU coprocessor on an M.2 2230 card. It adds 4 TOPS of acceleration for TensorFlow Lite models to a mini PC, industrial gateway or embedded board with an M.2 A+E slot. Ideal for smart video surveillance, object detection or local image classification, with a power consumption of only 2 W.
Highlights
- Google Edge TPU : 4 TOPS in int8, i.e. 2 TOPS per watt.
- Very fast on mobile models : Google quotes nearly 400 frames per second with MobileNet v2.
- M.2 2230 A+E key form factor, the same as Wi-Fi cards: it fits in the Wi-Fi slot of many mini PCs.
- Local processing : lower latency, no Internet connection required, private data.
- Extended temperature range : -20 to +85 °C in operation.
Technical specifications
| Features | Value |
|---|---|
| Coprocessor | Google Edge TPU |
| Performance | 4 TOPS (int8), 2 TOPS/W |
| Format | M.2 2230, A+E key (M.2-2230-A-E-S3) |
| Interface | PCIe Gen2 x1 |
| Dimensions and weights | 22 x 30 x 2.35 mm, 3.1 g |
| Power supply | 3,3 V ±10 % |
| Operating temperature | -20 à +85 °C |
| Supported systems | Linux Debian 10 or Ubuntu 16.04 and newer (x86-64 or ARMv8), Windows 10 64-bit (x86-64) |
Compatibility: read before ordering
- You need an M.2 A+E key slot connected via PCIe. An A+E key card does not fit in an M-key M.2 slot, the one used by NVMe SSDs.
- No documented Raspberry Pi compatibility : the M.2 HATs and NVMe adapters for Raspberry Pi 5 use the M key. Google does not document any Raspberry Pi installation for this version.
- Driver : the module works with the Edge TPU runtime and the Coral PCIe driver; the host must support MSI-X interrupts.
- Slot power : Google reports current peaks of up to 3 A depending on the model being run; the host must be able to handle them.
On Raspberry Pi: the right alternatives
- L'Google Coral USB Accelerator carries the same Edge TPU and plugs into a USB 3.0 port on any Raspberry Pi 4 or 5.
- On Raspberry Pi 5, theAI HAT+ (Hailo-8L or Hailo-8, 13 or 26 TOPS) connects to the PCIe port and integrates with Raspberry Pi OS.
- For local generative AI, theAI HAT+ 2 adds 40 TOPS (INT4) and 8 GB of dedicated memory.
Uses and projects
- Smart video surveillance : Frigate, which integrates with Home Assistant, supports the Coral in M.2 PCIe format to detect people and vehicles locally.
- Industrial gateways : quality control, counting, scene analysis without the cloud.
- Connected devices and embedded AI prototypes on mini PCs.
- To compare acceleration solutions, read our feature on AI HATs for Raspberry Pi 5.
Frequently asked questions
What is the difference between the Google Coral M.2 and the Coral USB?
Same 4 TOPS Edge TPU chip. The M.2 A+E installs in an internal M.2 PCIe slot; the Coral USB plugs into a USB port, which makes it compatible with Raspberry Pi boards.
Does the Coral M.2 A+E work on the Raspberry Pi 5 M.2 HAT?
No: the official M.2 HAT uses an M key, and the Coral card uses an A+E key. No Raspberry Pi compatibility is documented by Google.
Which AI models can it run?
TensorFlow Lite models quantized to int8 and compiled for the Edge TPU, for example for image classification or object detection.
Does it work with Frigate?
Yes, Frigate supports the PCIe and M.2 Coral as an object detector.
Which operating systems does it run on?
64-bit Linux (Debian 10, Ubuntu 16.04 or newer, x86-64 or ARMv8) and Windows 10 64-bit on x86-64.
Order with confidence on MONRASPBERRY : shipped within 24/48 h after receipt of payment from our stock based in France, free pickup point delivery on orders of €200 or more, official and tested products, returns within 14 days (with return authorization) and payment by Visa, Mastercard, PayPal, Apple Pay or Google Pay.
Additional information
| Accessory type | Modules |
|---|
In the same category...
