How to install Cuda and Cudnn for Deep Learning

By • July 04, 2025

✅ How I Successfully Installed TensorFlow with GPU Support on Windows (2025 Guide)

If you’ve ever struggled to get TensorFlow to detect your NVIDIA GPU on Windows — you’re not alone. I went through the same headache, but after fixing a few key mistakes, it finally worked. Here’s exactly what I did, step by step, including all the commands and setup process.

🎯 Goal

I wanted to set up TensorFlow 2.10 with GPU acceleration on my Windows laptop (RTX 4050 + i7 12th Gen). While regular CPU-based tasks were fine, I needed GPU power for deep learning. Unfortunately, it didn’t “just work” — and here’s what fixed it.

❌ The Mistake I Made

Like many beginners, I installed CUDA 10.1 (the version I found online), assuming it would work. But TensorFlow is very specific about what versions it supports. I was using:

tensorflow 2.10.0
CUDA 10.1 ❌ (wrong!)

But for GPU to work, TensorFlow 2.10 requires:

• CUDA 11.2

• cuDNN 8.1

🔧 Fixing the Setup — Step by Step

✅ 1. Uninstall Old CUDA

I couldn’t find CUDA 10.1 in my installed apps, so I deleted the folder manually:

C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.1

I also cleaned up old CUDA entries from my environment variables (more below).

✅ 2. Install CUDA 11.2

Download from the official archive:
👉 CUDA 11.2.2 Download Link

Follow the installer, and make sure it's installed to:

C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2

✅ 3. Install cuDNN 8.1

Download from the cuDNN archive (you’ll need an NVIDIA login):
👉 cuDNN v8.1.1 for CUDA 11.2

Unzip the files and copy them into the CUDA folder:

✅ I confirmed files like cudart64_110.dll and cudnn64_8.dll were present.

✅ 4. Set Environment Variables

Go to:

Start → Environment Variables → Edit the system environment variables

In the System Path, I added:

C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2\bin
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2\libnvvp

And created a new User variable:

CUDA_PATH = C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2

🧹 I also made sure no old CUDA (10.1) paths were left in the PATH.

🔄 5. Reboot Your System

This is critical. Skipping the reboot kept my old CUDA settings active. After reboot, the new ones took effect.

🐍 Installing TensorFlow (Correct Way)

After CUDA and cuDNN were set up, I created a virtual environment for TensorFlow:

# Create and activate virtual environment
python -m venv tf_gpu
.\tf_gpu\Scripts\activate

# Install TensorFlow 2.10 (GPU support built-in)
pip install tensorflow==2.10.0

✅ Note: TensorFlow 2.10 is the last version that supports native GPU on Windows.

✅ Verifying GPU Support in Python

Now comes the fun part — testing if it works:

import tensorflow as tf
print("TensorFlow version:", tf.__version__)
print("GPU available:", tf.config.list_physical_devices('GPU'))

Output:

TensorFlow version: 2.10.0
GPU available: [PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]

🔍 Optional: Check All Devices

from tensorflow.python.client import device_lib
print(device_lib.list_local_devices())

🧪 Bonus: Force TensorFlow to Use GPU

tf.debugging.set_log_device_placement(True)

with tf.device('/GPU:0'):
    a = tf.constant([[1.0, 2.0], [3.0, 4.0]])
    b = tf.constant([[1.0, 1.0], [0.0, 1.0]])
    c = tf.matmul(a, b)

print(c)

🧠 What I Learned

• TensorFlow is very strict about CUDA/cuDNN versions.

• Even if the files are installed correctly, old paths in the environment variables can silently break everything.

• Rebooting is non-negotiable — DLLs don’t refresh until a restart.

• The TensorFlow + GPU setup on Windows works perfectly, but only with the exact supported versions.

🔗 Useful Links

• TensorFlow GPU Setup Guide

• CUDA 11.2.2 Download

• cuDNN v8.1 Archive