0022 MPSTPU(M.2 M) EN

关键词

树莓派5、PCIe、TPU、DTPU、驱动安装、操作演示、AI、google、Coral、2230、2242、2280、NVMe SSD、固态硬盘 X1

一、简介

Coral M.2加速器是一款M.2尺寸的模组，可以通过PCIE接口为树莓派等平台扩展边缘TPU协处理器。我司的MPSTPU扩展板通过PCIe Switch扩展出了两路PCIe接口，其中一路为M.2 M-key接口，可外接M-Key接口的TPU模组；另一路为M.2 M key接口，可外接NVMe SSD固态硬盘。MPSTPU扩展板可在加入TPU边缘计算能力的同时为系统提供SSD海量存储。

二、硬件资源

1. 利用树莓派5的PCIe接口，采用PCie Switch芯片，将PCIe一扩二；采用0.5mm间距16Pin PCIe2.0 x1接口连接；

2. 引出一路PCIE M.2 M-key接口,支持NVMe协议2230/2242/2280尺寸固态硬盘(默认焊接2280固定柱)；注意：不支持NGFF和SATA协议的盘；

3. 引出一路PCIE M.2 M-key，支持M key接口的Google Coral Edge TPU；

4. 引出一路5V电源指示灯("PWR")，一路硬盘工作状态指示灯；

5. 设计采用1.5A高效DC-DC电路，可支持绝大部分固态盘（受限于x1接口，固态盘的实际峰值功耗仅为额定功耗的1/3）；

6. 扩展板采用内凹开槽设计，40Pin上方无遮挡不影响杜邦线连接；

7. 四个M2.5固定孔，和树莓派5定位孔吻合；

8. 沉金工艺，无铅生产；PCB板符合UL和RoHS认证,防火等级94V-0；

9. 可选配铝合金外壳。

三、系统烧写及设置

3.1 概述

本文档采用树莓派OS、Ubuntu系统和OpenWrt系统进行测试。

1) 树莓派OS的版本为2024-07-04-raspios-bookworm-arm64.img.xz，树莓派OS下载地址：

https://www.raspberrypi.com/software/operating-systems/#raspberry-pi-os-64-bit

2) Ubuntu系统的版本为ubuntu-24.04-preinstalled-desktop-arm64+raspi.img.xz，Ubuntu OS下载地址：

https://ubuntu.com/download/raspberry-pi

3) OpenWrt系统版本：

openwrt-bcm27xx-bcm2712-rpi-5-squashfs-sysupgrade-linux-6.1.100-20240805.img.gz

3.2 系统烧写在SD（TF）卡上

点击直达烧写方法说明。

3.3 系统烧写在SSD硬盘上

点击直达烧写方法说明。

IV. Run the TPU module on the Raspberry Pi OS

4.1 Run the TPU module

This chapter introduces how to run the TPU module on the Raspberry Pi OS.

The steps in this chapter apply to both systems booting from TF card and from SSD.

4.1.1 Config config.txt

After the system boots, open the Raspberry Pi terminal and enter the command:

sudo nano /boot/firmware/config.txt

Or:

sudo mousepad /boot/firmware/config.txt

Add the codes in the ending of this file:

dtparam=pciex1

kernel=kernel8.img

dtoverlay=pineboards-hat-ai

Then, save the file and reboot the system:

sudo reboot

4.1.2 Ensuring software environment

After reboot the system, open the Raspberry Pi terminal and enter the command:

uname -r

After ensuring the kernel version is greater than 4.18, enter:

lsmod | grep apex

Ensure there is no output, then you can begin installing the TPU driver.

4.1.3 Install necessary software

Ensure the network is connected to internet connection to the regions out of China (you may need to find your own method), and then add the Google TPU software library:

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 -

Update the software list after adding:

sudo apt-get update

Install the necessary software after the update is completed:

sudo apt-get install cmake libedgetpu1-std devscripts debhelper dkms dh-dkms

4.1.4 Install Gasket Driver

Enter the following commands in sequence to install the Gasket Driver:

git clone https://github.com/google/gasket-driver.git

cd gasket-driver

sudo debuild -us -uc -tc -b

cd ..

sudo dpkg -i gasket-dkms_1.0-18_all.deb

After the installation is complete, enter the following command to ensure that the driver and software were installed successfully:

sudo apt-get install gasket-dkms libedgetpu1-std

Then, add a udev rule to obtain device operation permissions:

sudo sh -c "echo 'SUBSYSTEM==\"apex\", MODE=\"0660\", GROUP=\"apex\"' >> /etc/udev/rules.d/65-apex.rules"

add user to apex:

sudo groupadd apex

sudo adduser $USER apex

Reboot the system after the setup is complete:

sudo reboot

4.1.5 Verification module testing and driver installation

After system starting, verification module testing and driver installation:

lspci -nn | grep 089a

Here, you need to see the TPU module, as shown in the following figure (specific parameter display is subject to the actual device):

ls /dev/apex_0

This board has a single TPU onboard, so you can see a TPU module:

By entering sudo lspci -v, you can see the driver loading, this must include the TPU driver, and if an SSD is inserted, it also needs to include the SSD module, as shown in the following figure:

4.1.6 Configuration of the Code Execution Environment

Running Google TPU module code requires Python versions between 3.6 and 3.9, while the latest Raspberry Pi system is Python 3.11. Therefore, we need to use Pyenv to download an older version of Python.

First, install the depends:

sudo apt-get install -y make build-essential libssl-dev zlib1g-dev \libbz2-dev libreadline-dev libsqlite3-dev wget curl llvm libncurses5-dev \libncursesw5-dev xz-utils tk-dev libffi-dev liblzma-dev python3-openssl

Install pyenv after the depends are installed:

curl https://pyenv.run | bash

Enter the command:

sudo nano ~/.bashrc

Or:

sudo mousepad ~/.bashrc

Add the codes in the ending of this file:

export PYENV_ROOT="$HOME/.pyenv"

[[ -d $PYENV_ROOT/bin ]] && export PATH="$PYENV_ROOT/bin:$PATH"

eval "$(pyenv init -)"

Save and exit after adding, then reload the shell:

exec "$SHELL"

Then, we can use pyenv to install an older version of Python:

pyenv install -v 3.8.8

After the installation is successful, we will change the system Python version to 3.8.8:

pyenv global 3.8.8

After the change is successful, you can check the current Python version is 3.8.8:

pyenv versions

python3 -V

4.1.7 Download the code and run the module

Enter the following commands in sequence to download:

mkdir coral && cd coral

git clone https://github.com/google-coral/pycoral.git

cd pycoral

cd test_data

git clone https://github.com/google-coral/test_data.git

Then, download the code and run the module:

pip3 install numpy

pip3 install Pillow

pip3 install --extra-index-url https://google-coral.github.io/py-repo/ pycoral

4.1.8 Run the code

After the module download is complete, back to the coral/pycoral path:

cd ..

Enter the following code to run it:

Note: You can copy the following code, but please paste it into a text editor first. Then, following the format shown above, remove any extra enter. After that, copy it again and paste it into the terminal to run.

python3 examples/classify_image.py \

--model test_data/test_data/mobilenet_v2_1.0_224_inat_bird_quant_edgetpu.tflite \

--labels test_data/test_data/inat_bird_labels.txt \

--input test_data/test_data/parrot.jpg