Thunderbolt 3 eGPU Bumblebee 方案尝鲜

最近从公司淘了一块 GTX 1060，淘宝了一个显卡盒折腾 TB3 eGPU 方案。由于笔记本经常带出门，打算采用即插即用的 bumblebee 方案（回家插上盒子用独显，拔下盒子可以带出门，可以正常待机不用重启）。下面记录一下 Arch 上配置的步骤和遇到的坑。

1、准备软件包

# pacman -S bumblebee primus lib32-primus nvidia-dkms lib32-nvidia-utils bolt

1	# pacman -S bumblebee primus lib32-primus nvidia-dkms lib32-nvidia-utils bolt

2、修改 bumblebee 配置

修改 /etc/bumblebee/xorg.conf.nvidia，在 Section “Device” 中加入：

Option "AllowExternalGpus" "true"

1	Option "AllowExternalGpus" "true"

修改 /etc/bumblebee/bumblebee.conf，在 [optirun] 中修改：

Bridge=primus

1	Bridge=primus

3、插入 TB 线，验证设备（不建议完全关闭设备验证！）

输入 boltctl 查看设备信息，记下设备的 uuid。

$ boltctl
 ● TUL TBX-550CA
   ├─ type:          peripheral
   ├─ name:          TBX-550CA
   ├─ vendor:        TUL
   ├─ uuid:          00xxxxxx-xxxx-xxxx-ffff-ffffffffffff
   ├─ status:        authorized
   │  ├─ domain:     domain0
   │  └─ authflags:  none
   ├─ authorized:    Fri 28 Sep 2018 08:23:25 AM UTC
   ├─ connected:     Fri 28 Sep 2018 08:23:21 AM UTC
   └─ stored:        Thu 27 Sep 2018 05:22:07 AM UTC
      ├─ policy:     auto
      └─ key:        no

$ boltctl

● TUL TBX-550CA

├─ type: peripheral

├─ name: TBX-550CA

├─ vendor: TUL

├─ uuid: 00xxxxxx-xxxx-xxxx-ffff-ffffffffffff

├─ status: authorized

│ ├─ domain: domain0

│ └─ authflags: none

├─ authorized: Fri 28 Sep 2018 08:23:25 AM UTC

├─ connected: Fri 28 Sep 2018 08:23:21 AM UTC

└─ stored: Thu 27 Sep 2018 05:22:07 AM UTC

├─ policy: auto

└─ key: no

验证设备并设置为自动验证：

$ boltctl enroll --policy=auto 00xxxxxx-xxxx-xxxx-ffff-ffffffffffff

1	$ boltctl enroll --policy=auto 00xxxxxx-xxxx-xxxx-ffff-ffffffffffff

4、启动 bumblebeed

# systemctl start bumblebeed

1	# systemctl start bumblebeed

5、测试、运行程序

$ optirun glxspheres64
Polygons in scene: 62464 (61 spheres * 1024 polys/spheres)
Visual ID of window: 0x13f
Context is Direct
OpenGL Renderer: GeForce GTX 1060 5GB/PCIe/SSE2
……

$ optirun glxspheres64

Polygons in scene: 62464 (61 spheres * 1024 polys/spheres)

Visual ID of window: 0x13f

Context is Direct

OpenGL Renderer: GeForce GTX 1060 5GB/PCIe/SSE2

……

6、安全删除设备（参考了 jpamills 博客里的脚本）

#!/bin/bash

secs=5
tbt_chain=/sys/bus/thunderbolt/devices/0-0/../../../..

echo "Unplug eGPU script started."
if [ "$(id -u)" != "0" ]; then
        echo "Please run using sudo. Exiting."
        exit 1
fi

systemctl restart bumblebeed
sleep 2
modprobe -r nvidia_modeset
modprobe -r nvidia-uvm
modprobe -r nvidia

if [ -e $tbt_chain/remove ]
then
        echo 1 > $tbt_chain/remove
        echo "Thunderbolt chain removed from PCI tree. Please unplug eGPU now."
        while [ $secs -gt 0 ]; do
                echo -ne "$secs to rescan...\033[0K\r"
                sleep 1
                : $((secs--))
        done
        echo 1 > /sys/bus/pci/rescan
        echo "Rescanned the PCI bus. Completed."
        exit 0
else
        echo "eGPU does not appear to be attached. Exiting."
        exit 1
fi

#!/bin/bash

secs=5

tbt_chain=/sys/bus/thunderbolt/devices/0-0/../../../..

echo "Unplug eGPU script started."

if [ "$(id -u)" != "0" ]; then

echo "Please run using sudo. Exiting."

exit 1

systemctl restart bumblebeed

sleep 2

modprobe -r nvidia_modeset

modprobe -r nvidia-uvm

modprobe -r nvidia

if [ -e $tbt_chain/remove ]

then

echo 1 > $tbt_chain/remove

echo "Thunderbolt chain removed from PCI tree. Please unplug eGPU now."

while [ $secs -gt 0 ]; do

echo -ne "$secs to rescan...\033[0K\r"

sleep 1

: $((secs--))

done

echo 1 > /sys/bus/pci/rescan

echo "Rescanned the PCI bus. Completed."

exit 0

else

echo "eGPU does not appear to be attached. Exiting."

exit 1

如果和我一样在盒子上插了外置硬盘，还可以在脚本里加上相应的 umount 语句。删除设备时需要先把使用设备的程序退出。

效果展示

剩下的问题

脚本里可以考虑通过判断 nvidia-smi 列出的 PID 自动杀掉所有还在用卡的进程
Vulkan 应用程序暂时无法使用外置卡运行（仍然使用集显），可能需要 bumblebee 提供支持