






🚀 Unlock lightning-fast local AI inferencing—because your projects deserve the edge!
The Seeed Studio Google Coral USB Accelerator is a compact, low-power USB 3.0 device featuring Google's Edge TPU ASIC, designed to accelerate TensorFlow Lite machine learning inferencing on Linux and Raspberry Pi systems. It drastically reduces inference latency and CPU load, supports popular ML architectures, and integrates seamlessly with Google Cloud, making it an essential tool for efficient, scalable edge AI development.

| ASIN | B084TBYKL9 |
| Are Batteries Included | No |
| Best Sellers Rank | 105,372 in Computers & Accessories ( See Top 100 in Computers & Accessories ) 370 in Barebone PCs |
| Brand | seeed studio |
| Customer Reviews | 4.2 4.2 out of 5 stars (41) |
| Date First Available | 14 Feb. 2020 |
| Guaranteed software updates until | unknown |
| Item Weight | 90 g |
| Item model number | 114991790U |
| Manufacturer | seeed studio |
| Memory Technology | LPDDR3 |
| Operating System | Linux |
| Package Dimensions | 13.6 x 10.3 x 3.2 cm; 90 g |
| Processor Brand | Microchip Technology |
| Processor Count | 1 |
| Processor Speed | 32 MHz |
| Processor Type | 80386 |
| Series | Studio |
| Wireless Type | Bluetooth |
J**N
Great addition for Frigate
Having purchased a Coral USB from a UK supplier that was DOA, it was with a little bit of trepidation that I ordered via Amazon's Global Store. However, the Amazon device worked out of the box and is a fabulous addition to a Frigate installation. My mini-PC's fan is no longer on 24/7. Inference is down from above 40ms to below 9ms. CPU usage is down from high 40% to below 12%. Seeing how difficult these are to get hold of, I might buy another, just in case.
E**K
Useful for TensorFlow Lite.
Unlike other USB sticks (like the Intel NCS2) this will only run TF Lite, nothing else. Works well for inference (cannot strictly train on this except under limited circumstances i.e. the FC/Dense layers in a model, all other weights must be frozen). If you are looking at developing for upcoming Android platform ASIC's (on a phone for instance) this is an ideal test device to make sure your models are ready and deployable. Believe folk buy for the Raspberry Pi... never had one but from what I hear very good for this as well. Very small and light, took me 20 minutes from unboxing to running inference demos under Ubuntu 19.04.
Z**N
Fast delivery and product is as described. Works really well with Frigate!
K**Y
Je l'utilise avec Frigate.Ca fait parfaitement le job avec 6 caméras.
J**2
Awesome. Helps me with my project. Have to tinker a bit with the command line on Linux, which ai helps with. A great hobbyist tool. Happy with the purchase
B**E
ma nessuno lo vendeva su amazon.... per darvi un'idea, in un video 640x480, l'algoritmo object detection (SSD Mobilenet) allenato su COCO dataset impiega 2 millisecondi a frame, indipendentemente dal numero di oggetti presenti... pazzesco, se pensate che RPI4 fa a malapena 3 frame al secondo. chiaramente tira da bestia se usato su usb3, e senza aumentarne il clock (si può fare per farlo andare ancora più veloce). il riconoscimento dei visi è spettacolare e velocissimo.
D**.
If you use Frigate in Home Assistant, this is a must-have. Remember it wont work with Frigate-FA - took me a bit extra time to configure mine as I was using the full access version of frigate, then installed the standard, and it worked like a charm.
Trustpilot
3 weeks ago
1 month ago