Will the Google Edge TPU 'Google Edge TPU' finally be released? What price are you interested in?

Tensor Processing Unit (TPU) is a customized ASIC for efficient machine learning using TensorFlow, which Google has been developing for a long time . " Google Edge TPU " which specialized TPU for edge devices such as IoT seems to be released at last.

Edge TPU Devices

Google Edge TPU Coral Dev Board with a removable SOM - Seeed Studio

To date, Google has focused on first-generation TPUs dedicated to reasoning on the cloud side ... ...

It is obvious that Google has made secret backwards a few years ago for "TPU" dedicated to making machine learning explosive - GIGAZINE

2nd Generation TPU possible up to the "training" processing of the neural network which was impossible with the 1st generation TPU.

2nd generation of Google's machine learning machine "TPU", 1 board 180 TFLOPS reaching 11.5 PFLOPS at 64 grids - GIGAZINE

Furthermore, it has about 8 times faster than the 2nd Generation TPU, and has announced third generation TPU which introduced liquid cooling system.

Google announced the third-generation processor dedicated to machine learning "TPU 3.0", so that cooling can not catch up and introduced liquid cooling system - GIGAZINE

The TPU that Google announced so far was primarily intended for use on the cloud service side, but as a new TPU for edge devices, Google added Google Edge TPU to Google, which was held in July 2018 We announce it in Cloud related developer event "Google Cloud Next '18". As its name suggests, Google Edge TPU is a TPU developed for use with edge devices such as IoT and Google as "having high power efficient inference performance of 4TOPS (Trillion Operations Per Second) / 2 Wh" There was.

SoM equipped with such Google Edge TPU was sold on Seeed Studio dealing with IoT products. The Google Edge TPU-equipped SoM is said to have made advance reservations for the release on 20th March 2019, and the selling price is 149.99 dollars (about 17,000 yen), but it is already sold out at the time of article creation It was.

Google Edge TPU Coral Dev Board with a removable SOM - Seeed Studio

According to Google, Google Edge TPU provides high performance ML (Machine Learning) reasoning for low-power devices, for example a state-of-the-art mobile vision model like MobileNet V2, in a power efficient manner and 100 fps or more It will be able to be executed with.

The detailed specifications of SoM and baseboard with Google Edge TPU are as follows.

◆ Google Edge TPU Module (SOM) specification
CPU: NXP i.MX 8 M SOC (Quad Cortex-A 53, Cortex-M 4 F)
GPU: Integrated GC 7000 Lite Graphics
ML Accelerator: Google Edge TPU Coprocessor
RAM (memory): 1 GB LPDDR 4
Flash memory: 8GB eMMC
Wireless: Wi-Fi 2 × 2 MIMO (802.11 b / g / n / ac 2.4 / 5 GHz), Bluetooth 4.1
Dimension: 40 mm × 48 mm

◆ Baseboard Specification
Flash memory: MicroSD slot
USB: Type-C OTG, Type-C power supply, Type-A 3.0 host, Micro-B serial console
LAN: Gigabit Ethernet port
Audio: 3.5 mm audio jack (CTIA compliant), digital PDM microphone (2), 2.54 mm 4 pin terminal for stereo speaker
Video: HDMI 2.0a (full size), 39 pin FFC connector (4 lanes) for MIPI-DSI display, 24 pin FFC connector (4 lanes) for MIPI-
GPIO: 40-pin extension header
Power: 5 V DC (USB Type-C)
Dimension: 85 mm × 56 mm

Support OS: Debian Linux
Support Framework: TensorFlow Lite

In addition, although it is a sales page on Seeed Studio that was visible at the beginning of the article creation, it is not available for viewing at the time of publishing the article. On the Google Edge TPU development board, the status of SoM with Google Edge TPU remains "appearing soon", so there is a good possibility that Seeed Studio has flown the release timing .

You can check the sales page at the timing that was available for viewing with Archive.today .

Google Edge TPU Coral Dev Board with a removable SOM - Seeed Studio

in Hardware, Posted by logu_ii