linux-docs

egpu-linux-install

Installing external GPU enclosure for use on laptop running Linux distribution; utilizing hertg’s egpu-switcher script to facilitate

Prerequisites

• a Thunderbolt 3-compatible (or newer) laptop PC running a Linux desktop operating system based on one of the following: Ubuntu, Fedora, Arch (in my case, we will be using Pop!_OS 22.04 LTS)
• sudo or administrator privileges
• an internet connection
• a Thunderbolt 3-compatible (or newer) external GPU enclosure with discrete graphics card (in my case, I will be using a Cooler Master EG200 with an RTX 2060)

1. Install dependencies for script

• hertg’s script requires the Go toolchain to be installed for building from source; open your terminal and enter the commands as follows:
  • clone the Go source code: wget https://go.dev/dl/go1.20.2.linux-amd64.tar.gz
  • remove any previous installations of Go: sudo rm -rf /usr/local/go
  • extract the Go source code to our runtime directory: sudo tar -C /usr/local -xzf ./go1.20.2.linux-amd64.tar.gz (your archive tar name may differ from mine; at the time of writing, 1.20.2 is the latest stable version of the Go toolchain)
• Add the Go binary to your PATH:
  • for one-time use: export PATH=$PATH:/usr/local/go/bin
  • for permanent or repeated use, edit your .bashrc (.zshrc in my case) in your home directory (/home/user/), and add the above export line at the very end.
• Either close and re-open your terminal, or issue the following command: source .bashrc
• Verify your Go installation with the following command: go version
  • the output should match the following: go version go1.20.2 linux/amd64 (again, your version number may differ)
• To cover the bases, install your regular build dependencies with the following command: sudo apt install build-essential gcc make (you may receive output that these packages are already installed, if so skip to step 2)
• If you are on Pop!_OS, you may already have the necessary Nvidia drivers depending on which image you downloaded when you installed the OS on your laptop; confirm you have the necessary Nvidia drivers with the following command: nvidia-smi
  • if you receive a message about “command not found”, install the necessary nvidia drivers with the following command and reboot once completed: sudo apt install system76-driver-nvidia
  • if you are not running Pop!_OS, refer to this link for your distribution’s specific instructions for installing: System76 Driver (Install) - System76 Support
• If you are on Pop!_OS, you already have the necessary system76-power package to enable graphics switching; if you are on stock Ubuntu or a derivative, Fedora, or an Arch-based distribution, follow the steps below based on your specific needs:
  • Ubuntu: Graphics Switching (Ubuntu) - System76 Support
  • Fedora: System76 Software (Install - Other Operating Systems) - System76 Support
  • Arch: System76 Software (Install - Other Operating Systems) - System76 Support

2. Clone script source and prepare build environment

• Create a “temp” directory to clone the source code for hertg’s egpu-switcher script; in my case, we will create a “.tmp” folder in our home directory: mkdir ~/.tmp
• Change directories into “.tmp”: cd ~/.tmp
• Clone the source code from GitHub: git clone https://github.com/hertg/egpu-switcher
• Change directories into the source code folder: cd egpu-switcher

3. Install and configure the script, and test

• Build the installation: make build -s
  • you will see multiple lines about “go” downloading dependencies; this is fine as long as the last line of output matches: binary compiled => ./bin/egpu-switcher
• Install the script: sudo make install -s
  • your output should match the following:

      binary installed at /usr/bin/egpu-switcher  
      docs generated => ./docs  
      manpages installed in /usr/share/man/man1
    

• At this point, go ahead and plug in your eGPU and power it on
• Confirm the eGPU is detected by observing both of the following:
  • your laptop now shows it is connected to power and charging;
  • “Settings > Privacy > Thunderbolt” shows the eGPU enclosure as “Authorized”
• If the eGPU is not automatically detected, unplug, reboot, and plug back in
• In your terminal, issue the following command: sudo egpu-switcher enable
  • if the steps up to this point have been followed, you should be presented with two options; enter the corresponding number for the nvidia option, then reboot your PC with the eGPU still plugged in
  • the egpu-switcher script creates the necessary symbolic links and a startup service for the laptop to recognize the GPU at boot when the GDM service is started.
• You should now be able to run nvidia-smi in your terminal and receive a snapshot of your eGPU’s current driver version, processing/memory load, and temperature.

4. Credits

• hertg, developer of the egpu-switcher script; you can follow his work on GitHub here: hertg (Michael Hertig) - GitHub
• S31bz, responsible for informing me of the Cooler Master EG200’s real-world Linux compatibility after his own configurations with an AMD GPU and Framework 11th Gen laptop (source); you can follow S31bz’s work on his website here: Alex Seibz’ Site

5. DISCLAIMERS - WIP

• This document is written with the assumption of an Nvidia GPU; steps may differ for AMD, which this guide will not cover.
• If you have an AMD GPU, your drivers are already baked into the mesa package provided by either the nvidia or intel Pop!_OS ISOs; no further driver configuration is needed.
• As of yet, there is NO hot-plug support for external graphics in Linux. You will need to reboot your laptop before you can utilize the eGPU. This is for both Nvidia AND AMD cards.
• There may be a slight perceived bottleneck on the GPU’s performance if pushed; Thunderbolt 3 exposes a maximum of 4 PCIe 3.0 lanes, and most GPUs work at x16-lane speeds in a regular desktop PC.