🚀 Use Your GPU Inside VS Code with WSL2 (Complete Guide)

It’s a really big issue that we can’t use GPU inside our beloved VS-Code and have to use Google Collab instead.
It’s time to use your RTX 4060s (or whatever GPU you have) to full extent inside VS Code.

From my experience I did everything that should’ve worked (CUDA, cuDNN, Python 3.11, VS Code, conda env) and TensorFlow still ran on CPU.
This is a very common pain point:

TensorFlow pip wheels on native Windows (>= 2.11) are CPU-only
Even a perfect CUDA/cuDNN install will NOT make Windows TensorFlow use GPU.

So the official + recommended way:

1. Run TensorFlow with GPU inside WSL2 Ubuntu
2. Linux wheels include CUDA + cuDNN automatically
3. Most ML/AI engineers on Windows use this setup

---

⭐ What is WSL?

WSL = Windows Subsystem for Linux, an official Microsoft technology.

It allows you to run a full Linux system inside Windows without dual boot.

Think of it like:

• Running an Android app on Windows
• Running Linux commands inside Windows
• With direct GPU access for deep learning

SAFE? → Yes, used by millions.
Mess files? → No, Ubuntu is isolated.

---

⚡ Quick WSL2 Setup (Admin PowerShell)

1️ Install WSL + Ubuntu

wsl --install -d ubuntu-22.04

What happens:

• Downloads Ubuntu 22.04
• Installs WSL2 backend
• Prepares Linux subsystem

If asked → Restart PC.

---

2️ First Launch of Ubuntu

You’ll see:

Installing, this may take a few minutes...

Then it will ask:

Enter new UNIX username:

Pick any username & password.

---

3️ Check GPU inside Ubuntu

nvidia-smi

If you see your RTX GPU → GPU works.
If not → driver issue.

---

🐍 Install Python + TensorFlow GPU Environment (Ubuntu)

Run all commands inside Ubuntu terminal.

---

Step 1 -- Update Ubuntu

sudo apt update && sudo apt upgrade -y

---

Step 2 -- Install Python, venv, pip

sudo apt install -y python3 python3-venv python3-pip build-essential

---

Step 3 -- Create Environment

python3 -m venv ML-DL-lat

---

Step 4 -- Activate Environment

source ML-DL-lat/bin/activate

To reactivate later:

source ~/ML-DL-lat/bin/activate

---

Install Full ML + DL Stack (GPU Enabled)

pip install --upgrade pip setuptools wheel

---

TensorFlow GPU

Linux wheels automatically include CUDA + cuDNN.

pip install "tensorflow[and-cuda]"

---

PyTorch GPU (CUDA 12.1)

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121

---

Core ML Libraries

pip install numpy pandas scikit-learn scipy matplotlib seaborn tqdm

---

NLP / HuggingFace

pip install transformers datasets accelerate evaluate sentencepiece

---

Computer Vision Libraries

pip install opencv-python pillow imageio

---

Jupyter + Tools

pip install jupyter jupyterlab ipywidgets

---

Verify GPU Support

python - << 'PY'
import tensorflow as tf
print("TensorFlow:", tf.__version__)
print("Built with CUDA:", tf.test.is_built_with_cuda())
print("GPUs:", tf.config.list_physical_devices("GPU"))
PY

Expected:

TensorFlow: 2.20.0
built_with_cuda: True
GPUs: [PhysicalDevice('/physical_device:GPU:0')]
PyTorch: 2.5.1+cu121
cuda: True
Device: NVIDIA GeForce RTX 4050 Laptop GPU

---

🎉 You Now Have:

• TensorFlow GPU
• PyTorch GPU
• Full ML + DL stack
• CUDA working inside WSL
• Clean isolated environment
• Setup recommended by NVIDIA, Microsoft, Google

---

Setup VS Code Integration

Check VS Code path:

Test-Path "$env:LOCALAPPDATA\Programs\Microsoft VS Code\Code.exe"
Test-Path "C:\Program Files\Microsoft VS Code\Code.exe"
Test-Path "C:\Program Files (x86)\Microsoft VS Code\Code.exe"

Lets say correct path is:

$env:LOCALAPPDATA\Programs\Microsoft VS Code\Code.exe

---

Add “Open in WSL (Ubuntu)” to Right-Click Menu

Paste ALL of this in ADMIN PowerShell:

# Path to VS Code (confirmed)
$codePath = "$env:LOCALAPPDATA\Programs\Microsoft VS Code\Code.exe"

# 1) Create helper script in C:\Windows
$scriptPath = "C:\Windows\open-in-wsl.ps1"

$scriptContent = @'
param([string]$winPath)

if ([string]::IsNullOrEmpty($winPath)) {
    $winPath = (Get-Location).ProviderPath
}

$winPath = $winPath.TrimEnd('\')

$drive = $winPath.Substring(0,1).ToLower()
$rest  = $winPath.Substring(2).Replace("\","/")
$wslPath = "/mnt/$drive/$rest"

$codeExe = "$env:LOCALAPPDATA\Programs\Microsoft VS Code\Code.exe"
$folderUri = "vscode-remote://wsl+Ubuntu-22.04$wslPath"

Start-Process -FilePath $codeExe -ArgumentList "--folder-uri", $folderUri
'@

Set-Content -Path $scriptPath -Value $scriptContent -Encoding UTF8 -Force
Write-Host "Created script: $scriptPath"

reg add "HKCR\Directory\shell\OpenInWSL" /ve /d "Open in WSL (Ubuntu)" /f
reg add "HKCR\Directory\shell\OpenInWSL" /v Icon /d "$codePath" /f

$cmd = 'powershell.exe -NoProfile -ExecutionPolicy Bypass -File "C:\Windows\open-in-wsl.ps1" "%V"'
reg add "HKCR\Directory\shell\OpenInWSL\command" /ve /d "$cmd" /f

Write-Host "*** Installed successfully! ***"

---

Fix Windows 11 “Show More Options”

reg add "HKCU\Software\Classes\CLSID\{86ca1aa0-34aa-4e8b-a509-50c905bae2a2}" /f
reg add "HKCU\Software\Classes\CLSID\{86ca1aa0-34aa-4e8b-a509-50c905bae2a2}\InprocServer32" /ve /f

Restart Explorer:

Stop-Process -Name explorer -Force

Then run:

explorer

---

Final Setup in VS Code

Right click on the folder you want to open and select "open in WSL ubuntu" and open a new terminal in VS code

Activate environment:

source ~/ML-DL-lat/bin/activate

Install Jupyter kernel:

python -m ipykernel install --user --name ML-DL-lat --display-name "Python (ML-DL-lat)"

Select kernel:

• Python (ML-DL-lat)

---

👨‍💻 Test GPU Inside VS Code

import tensorflow as tf
print(tf.__version__)
print(tf.config.list_physical_devices("GPU"))

Expected:

['/physical_device:GPU:0']