Component guide · Advanced · 3 min read
Google Coral Edge TPU

Coral Dev Board Mini, M.2 B+M Key, M.2 A+E Key, M.2 Dual Edge TPU and Mini PCIe Accelerators and KKSB Coral Dev Board Case
By Geekobot Learning Team · Updated 29 Oct 2026
On this page
Coral boards run TensorFlow Lite models on Google's Edge TPU at about 4 TOPS using only 2W. The M.2 and Mini PCIe cards add the TPU to a PC, mini PC or single-board computer with a matching slot; the Dual Edge TPU card has two. The Dev Board Mini is a small Linux computer with the TPU built in. A very popular use is object detection in the Frigate camera recorder.
Key specs at a glance
- Performance
- 4 TOPS per Edge TPU, about 2W
- Cards
- M.2 A+E, M.2 B+M, Mini PCIe, Dual Edge TPU (M.2 E-key)
- Dev Board Mini
- MediaTek 8167s, 2GB RAM, 8GB eMMC
- Models
- 8-bit quantised TensorFlow Lite compiled for Edge TPU
- OS
- Linux (Debian/Ubuntu), Windows drivers for PCIe
- Dual TPU
- Needs a slot wired for two PCIe lanes to use both
Models in this guide
These are alternatives. Pick the one that fits your project; you only need one.
Option 1Google Coral Dev Board Mini G650-03324-01 Edge TPU AI SBC₹15,569
Option 2Google Coral Edge TPU M.2 B+M ML Accelerator Module₹3,659
Option 3Google Coral M.2 Accelerator with Dual Edge TPU₹7,869
Option 4Google Coral M.2 Edge TPU Accelerator A+E Key₹4,789
Option 5Google Coral Mini PCIe Edge TPU ML Accelerator₹4,389
Option 6KKSB Cases Google Coral Dev Board Case – Black/Grey Aluminum₹1,539
What the Google Coral Edge TPU: Dev Board Mini, M.2 and Mini PCIe accelerators is
Google Coral Edge TPU accelerators, in USB, M.2 and mini PCIe forms and on the Coral Dev Board, run TensorFlow Lite models fast at the edge, for example object detection on a Raspberry Pi.
Check your slot key before buying: A+E keys fit Wi-Fi slots, B+M keys fit SSD-style slots. Many boards wire only one PCIe lane to an E-key slot, so the Dual Edge TPU then works as a single TPU.
Install the gasket and apex drivers and the Edge TPU runtime, then use models from the Coral model zoo or compile your own quantised TFLite model with the Edge TPU compiler. In Frigate, set the detector type to edgetpu and the device to pci.
Google Coral Edge TPU: Dev Board Mini, M.2 and Mini PCIe accelerators pinout
The pins on the Google Coral Edge TPU: Dev Board Mini, M.2 and Mini PCIe accelerators and what each one does.
| Pin | Function |
|---|---|
| M.2 / mPCIe edge | PCIe Gen2 x1 and power |
| USB-C (Dev Board Mini) | Power and OTG |
| 40-pin header | GPIO on Dev Board Mini |
| Camera | MIPI CSI on Dev Board Mini |
| Heatsink pad | Thermal contact |
- M.2 / mPCIe edge
- PCIe Gen2 x1 and power
- USB-C (Dev Board Mini)
- Power and OTG
- 40-pin header
- GPIO on Dev Board Mini
- Camera
- MIPI CSI on Dev Board Mini
- Heatsink pad
- Thermal contact
Wiring the Google Coral Edge TPU: Dev Board Mini, M.2 and Mini PCIe accelerators
Give the card airflow; the TPU slows down if it reaches its thermal limit.
| Google Coral Edge TPU: Dev Board Mini, M.2 and Mini PCIe accelerators pin | Connection | Note |
|---|---|---|
| M.2 card | Host M.2 slot | Match the key |
| Driver | gasket-dkms + libedgetpu | Then reboot |
| Frigate | detectors: edgetpu, device: pci | In config.yml |
- M.2 card
- Connection:
- Host M.2 slot
- Note:
- Match the key
- Driver
- Connection:
- gasket-dkms + libedgetpu
- Note:
- Then reboot
- Frigate
- Connection:
- detectors: edgetpu, device: pci
- Note:
- In config.yml
Example code
On newer kernels the gasket-dkms package may need building from the community-maintained source.
echo "deb https://packages.cloud.google.com/apt coral-edgetpu-stable main" | \
sudo tee /etc/apt/sources.list.d/coral-edgetpu.list
curl https://packages.cloud.google.com/apt/doc/apt-key.gpg | sudo apt-key add -
sudo apt update
sudo apt install -y gasket-dkms libedgetpu1-std
sudo reboot
# After reboot the TPU should appear:
ls /dev/apex_0Setup and tuning
install the Edge TPU runtime and PyCoral; check which Python and OS versions are supported, because newer ones may need older Python.
use models compiled for the Edge TPU (quantised int8 TFLite run through the Edge TPU compiler).
plug into a USB 3 port for full speed; USB 2 works but is slower.
install the gasket and apex drivers; check the slot key (A+E or B+M).
the TPU throttles when hot; give it airflow in continuous use.
What you need
Shop this list6 of 6 found in our store · ₹37,814
Google Coral Dev Board Mini G650-03324-01 Edge TPU AI SBC: the part this guide coversGoogle Coral Dev Board Mini G650-03324-01 Edge TPU AI SBC₹15,569
Google Coral Edge TPU M.2 B+M ML Accelerator Module: another version this guide coversGoogle Coral Edge TPU M.2 B+M ML Accelerator Module₹3,659
Google Coral M.2 Accelerator with Dual Edge TPU: another version this guide coversGoogle Coral M.2 Accelerator with Dual Edge TPU₹7,869
Google Google Coral M.2 Edge TPU Accelerator A+E Key: another version this guide coversGoogle Coral M.2 Edge TPU Accelerator A+E Key₹4,789
Google Coral Mini PCIe Edge TPU ML Accelerator: another version this guide coversGoogle Coral Mini PCIe Edge TPU ML Accelerator₹4,389
KKSB Cases Google Coral Dev Board Case – Black/Grey Aluminum: another version this guide coversKKSB Cases Google Coral Dev Board Case – Black/Grey Aluminum₹1,539
Alternatives and when to choose them
- Edge AI boards: all accelerators compared.
- Raspberry Pi AI HAT: Hailo, more TOPS.
- Jetson modules: GPU-based AI.
Common mistakes to avoid
- Running an uncompiled TFLite model and expecting TPU speed.
- Using a newer Python than PyCoral supports.
- Plugging the USB accelerator into a USB 2 port.
- Running it hot with no airflow.
Frequently asked questions
Can I use the Coral USB Accelerator with a Raspberry Pi?
Yes. Install the Edge TPU runtime and PyCoral, and use a USB 3 port on the Pi 4 or Pi 5.
Can it run any TensorFlow model?
No. Models must be quantised to int8 TFLite and compiled with the Edge TPU compiler.
Why is my model slower than expected?
Some layers may not be supported and run on the CPU; check the compiler log.
Is the Coral still supported?
It works, but software updates are slow, so newer Python versions may need workarounds.
How fast is the Edge TPU?
Around 4 TOPS, enough for real-time object detection with MobileNet SSD models.
Parts for this guide
Keep learning
Sources
Written by the Geekobot team from manufacturer datasheets. Always check the datasheet for the exact part and variant you buy.
