Absolutely, Docker on Windows 10 does support GPU pass-through—but the setup varies based on your Docker backend. The WSL 2 backend is the modern, user-friendly path (and what most people use these days), so we’ll focus on that first, with a note on the more complex Hyper-V alternative.

Prerequisites #

Before you start, make sure you meet these requirements:

• Windows 10 Version: You need at least Windows 10 2004 (Build 19041) or later. Check this by pressing Win + R, typing winver, and verifying the version number.
• Updated GPU Drivers:
  • For NVIDIA: Install the latest NVIDIA Driver for WSL (not the standard desktop driver—look for this specific package on NVIDIA’s website).
  • For AMD: Use the AMD Adrenalin software to update to a driver version that supports WSL GPU integration.
  • For Intel: Install the latest Intel graphics driver with WSL support.
• Docker Desktop with WSL 2 Enabled: Ensure Docker Desktop is set to use the WSL 2 backend (this is the default for new installs, but worth confirming).

Step-by-Step Setup (WSL 2 Backend) #

1. Enable WSL 2 (if not already done)

   • Open PowerShell as Administrator (right-click Start → Windows PowerShell (Admin)).
   • Run the command:

     wsl --install
     

   • Restart your computer when prompted. This installs Ubuntu as the default WSL 2 distro, but you can use any distro you prefer.

2. Configure Docker Desktop for WSL 2

   • Open Docker Desktop, click the gear icon (Settings) in the top-right corner.
   • Go to General and check the box for Use the WSL 2 based engine.
   • Navigate to Resources → WSL Integration and toggle on the WSL 2 distro you want to use (e.g., Ubuntu).
   • Click Apply & Restart to save changes.

3. Run a GPU-Aware Container

   • Open a command prompt or PowerShell window.
   • For NVIDIA GPUs, run this test command to verify GPU access:

     docker run --rm --gpus all nvidia/cuda:11.8.0-base-ubuntu22.04 nvidia-smi
     

   • For AMD GPUs, use this command instead:

     docker run --rm --gpus all rocm/pytorch:latest rocm-smi
     

   • For Intel GPUs, test with:

     docker run --rm --gpus all intel/oneapi-basekit:latest sycl-ls
     

   If everything works, you’ll see output showing your GPU model, driver version, and other details.

Troubleshooting Common Issues #

• "GPU not found" errors in Docker:
  • First, check if WSL 2 can access the GPU directly. Open your WSL 2 distro terminal and run nvidia-smi (NVIDIA) or rocm-smi (AMD). If this fails, your GPU driver isn’t properly set up for WSL 2—reinstall the WSL-specific driver.
  • Ensure Docker Desktop’s WSL Integration is enabled for your distro. Double-check the Settings menu.
• No WSL Integration option in Docker Desktop:
  • This means WSL 2 isn’t installed or configured correctly. Re-run wsl --install and make sure your Windows version meets the minimum requirement.
• Hyper-V conflicts: If you had Hyper-V enabled before switching to WSL 2, ensure Docker is set to use WSL 2 (not Hyper-V) in the General settings.

Hyper-V Mode Alternative (Advanced) #

If you must use the Hyper-V backend (not recommended for most users), you can enable GPU pass-through via Discrete Device Assignment (DDA). This requires:

• Disabling Secure Boot in your BIOS/UEFI.
• Uninstalling the GPU driver from Windows (so it can be assigned to the Hyper-V VM).
• Using PowerShell commands to attach the GPU to the Docker VM.
  This process is far more complex and can cause stability issues, so stick with WSL 2 unless you have a specific reason to use Hyper-V.

内容的提问来源于stack exchange，提问作者Srihari Humbarwadi