Enable GPU Usage Locally for Deep Learning

This is a step-by-step guide to enable GPU training for deep learning projects in any code editor, whether it's Jupyter Notebook, VS Code, or PyCharm Professional.

Note: This setup was personally tested on my university's lab computer, which has an NVIDIA RTX A6000 GPU.

---

Step 1: Download NVIDIA Video Driver

• Check your GPU model by running the following command in the terminal:

nvidia-smi

• Copy and paste your GPU name into the search bar on NVIDIA's driver download page.
• Download the latest driver for your GPU:
  • NVIDIA GPU Driver Download
• Open it and install with default settings.

---

Step 2: Install Visual Studio C++

• Download and install Visual Studio Community Edition.
• Open Visual Studio and update if prompted, then close it.
• Download Visual Studio

---

Step 3: Install Anaconda Navigator

• Install everything using the default settings.
• Download Anaconda

---

Step 4: Install CUDA Toolkit

• Run the following command to check your CUDA version:

nvidia-smi

• Visit PyTorch Local Installation Guide to find the correct CUDA version.
• Download the CUDA Toolkit for your version:
  • Windows -> Architecture: x86_64
  • Version: 11 (For windows 11)
  • Installer Type: .exe (local)
  • Download CUDA Toolkit

Verify CUDA Installation:

1. Go to C:\Program Files
2. Look for NVIDIA GPU Computing Toolkit folder
3. Inside, check for a CUDA folder
4. Open it and ensure it contains a folder named 12.x (or your installed version)

---

Step 5: Install cuDNN

• Download the latest cuDNN version.
• Choose Local Windows Installer (ZIP).
• Download cuDNN
• Open it and install with default settings.

---

Step 6: Copy & Paste cuDNN Files

Steps:

1. Open File Explorer and split the screen into two sections.
2. Navigate to:

   C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\12.x
   

3. Open the cuDNN archive in another window.
4. Copy and paste the contents of the following folders:
   • all contents of bin from cuDNN archive → Paste into CUDA\12.x\bin
   • all contents of include from cuDNN archive → Paste into CUDA\12.x\include
   • all contents of lib from cuDNN archive → Paste into CUDA\12.x\lib

Verify Environment Variables:

1. Open Edit the System Environment Variables via Windows Search.
2. Check if CUDA_PATH and CUDA_PATH_V12_x exist under system variables.
3. If they match your CUDA installation path, you're good to go!

---

Step 7: Install PyTorch

• Go to PyTorch Official Site
• Select the stable version.
• Choose the correct CUDA version.
• Copy the given pip command from website and paste it into your VS Code terminal:
• It looks like this: (It is for CUDA 12.6)

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu126

Using Git Bash in VS Code:

1. Open Anaconda Navigator and choose and open VS Code
2. Open VS Code Terminal
3. Change the terminal to Git Bash
4. Paste and execute the copied command

---

Step 8: Verify GPU Installation

Make a new .ipynb file Run the following script to check if your GPU is correctly set up:

import torch

print("Number of GPUs:", torch.cuda.device_count())
print("GPU Name:", torch.cuda.get_device_name(0))

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
print('Using device:', device)

Expected Output:

Number of GPUs: 1
GPU Name: NVIDIA RTX A6000
Using device: cuda

---

✅ Now, you are ready to train deep learning models using GPU! 🚀