06/07/23, 13:16 Setting up a Deep Learning Workplace with an NVIDIA Graphics Card (GPU) — for Windows OS | by Rukshan Pramoditha | D…

You have 1 free member-only story left this month. Sign up for Medium and get an extra one.

Member-only story

Setting up a Deep Learning Workplace with an

NVIDIA Graphics Card (GPU) — for Windows
OS
To run deep learning models incredibly faster with Keras and TensorFlow — Neural
Networks and Deep Learning Course: Part 12

Rukshan Pramoditha · Follow

Published in Data Science 365
10 min read · May 23, 2022

Listen Share

https://medium.com/data-science-365/setting-up-a-deep-learning-workplace-with-an-nvidia-graphics-card-gpu-for-windows-os-b6bff06eeec7 1/25
06/07/23, 13:16 Setting up a Deep Learning Workplace with an NVIDIA Graphics Card (GPU) — for Windows OS | by Rukshan Pramoditha | D…

Original photo by Sanni Sahil on Unsplash, edited by author

In our “Neural Networks and Deep Learning Course”, we’ll often use Keras and
TensorFlow libraries to do the programming part of deep learning models.

Before using any of these libraries, we need to set up our PC or laptop by installing
and configuring them on our system. Although these libraries can be installed on
different types of operating systems, today’s discussion is limited to the Windows
operating system. I also assume that you have a laptop or PC with the latest edition
of Windows installed.

Today, in this tutorial, we’ll discuss,

How to install Python from the Anaconda distribution

How to install Keras and TensorFlow libraries with both CPU and GPU support

The importance of running Keras and TensorFlow code on GPU

How to verify the installations so that you will not get any installation error

Keras and TensorFlow are Python-based libraries. Before installing any of them,
first, we need to install Python.

Installing Python on Windows

There are many ways to install Python on your PC or laptop. Because we are
performing ML and DL tasks, I highly recommend you install Python from the
Anaconda distribution — The world’s most popular open-source Python distribution
platform.

Step 1: Go to the Anaconda distribution website.

Step 2: Select the type of Anaconda installer that you’re looking for. For me, it’s
the 64-Bit Windows Graphical Installer.

https://medium.com/data-science-365/setting-up-a-deep-learning-workplace-with-an-nvidia-graphics-card-gpu-for-windows-os-b6bff06eeec7 2/25
06/07/23, 13:16 Setting up a Deep Learning Workplace with an NVIDIA Graphics Card (GPU) — for Windows OS | by Rukshan Pramoditha | D…

(Image from Anaconda website, screenshot by author)

Or the website will automatically detect the right installer for your PC or laptop as
follows.

(Image from Anaconda website, screenshot by author)

Step 3: Either way, download the right installer and save the .exe file.

Step 4: Run the .exe file and follow the on-screen instructions to install the
software. The important thing is to keep the recommended settings and default
selections when you install Anaconda. Only change them if you know what
you’re doing!

After completing the setup, Python, data science-related libraries such as NumPy,
Pandas, Matplotlib, Scikit-learn, etc and the Anaconda Navigator will be installed on
your PC.
https://medium.com/data-science-365/setting-up-a-deep-learning-workplace-with-an-nvidia-graphics-card-gpu-for-windows-os-b6bff06eeec7 3/25
06/07/23, 13:16 Setting up a Deep Learning Workplace with an NVIDIA Graphics Card (GPU) — for Windows OS | by Rukshan Pramoditha | D…

Step 5: To verify the Python installation, open the Anaconda Prompt, type
python and hit enter.

(Image by author)

This will return the Python version. In my case, it is 3.9.12.

Installing Kears and TensorFlow on Windows

Libraries like Keras and TensorFlow do not come with the standard installation of
Anaconda. Therefore, we need to install them separately.

Step 1: Open the Anaconda Navigator — A GUI included in Anaconda that allows
you to easily manage packages (libraries), environments, IDEs, etc.

Step 2: Go to Environments tab → base (root) → Open Terminal. A command

window will appear.

(Image by author)

Step 3: Type the following command and hit enter to install both Keras and
TensorFlow. Since TensorFlow includes Keras, it’s not necessary to install Keras

https://medium.com/data-science-365/setting-up-a-deep-learning-workplace-with-an-nvidia-graphics-card-gpu-for-windows-os-b6bff06eeec7 4/25
06/07/23, 13:16 Setting up a Deep Learning Workplace with an NVIDIA Graphics Card (GPU) — for Windows OS | by Rukshan Pramoditha | D…

separately. We only install TensorFlow and import Keras from TensorFlow! This
is the most recommend method when working with these two libraries.

pip install tensorflow

(Image by author)

The above command will install the current stable release of TensorFlow and Keras
with both CPU and GPU support. As in older versions of TensorFlow, we do not need
to install separate packages for CPU and GPU. The pip installation includes both
CPU and GPU support but requires additional drivers and software to be configured
later (more on this shortly).

Step 4: To verify the installation, open a Python 3 Jupyter Notebook from the
Anaconda Navigator and execute the following commands.

import tensorflow as tf #Import TensorFlow

from tensorflow import keras #Import Keras from TensorFlow

If you get no error, you’ve successfully, installed TensorFlow. So, congratulations!

To get the TensorFlow version, run the following two lines of code.

import tensorflow as tf
tf.__version__

This returns 2.9.0, in my case.

https://medium.com/data-science-365/setting-up-a-deep-learning-workplace-with-an-nvidia-graphics-card-gpu-for-windows-os-b6bff06eeec7 5/25
06/07/23, 13:16 Setting up a Deep Learning Workplace with an NVIDIA Graphics Card (GPU) — for Windows OS | by Rukshan Pramoditha | D…

To get GPU support with Keras and TensorFlow, we need to do some additional
configurations on our system. We need to download and install the following
software.

NVIDIA GPU drivers

Microsoft Visual Studio (Prerequisite for CUDA Toolkit)

CUDA Toolkit

cuDNN SDK

By default, Keras and TensorFlow are looking for a GPU and if it is not available or
not configured, they will run on the CPU.

Only the NVIDIA GPUs are supported with TensorFlow and Keras. I repeat this. Only
the NVIDIA GPUs! Your NVIDIA GPU should also be CUDA-capable (more on this
shortly).

If you do not have an NVIDIA GPU, unfortunately, you cannot get GPU support with
Keras and TensorFlow. Therefore, you do not need to follow the below steps.
Alternatively, you can use Colab's free GPU!

The importance of running Keras and TensorFlow code on GPU

Here is the text directly taken from my “Why GPUs for Machine Learning and Deep
Learning?” article.

Some deep learning models and tensors require a lot of computational power to execute.
This is because they have a vast amount of data and thousands of parameters. A CPU
cannot handle these types of resource-intensive tasks efficiently.

As data and model parameters increase, more computational power is required to train
the models. You should expect higher runtime for those models if you only use the CPU.

To overcome this, high-performance GPUs are available at an affordable cost. GPUs are
ideal for doing computations that involve a lot of data. GPUs are capable of executing
operations with lower runtime thanks to their higher number of cores.

Configuring the system to get GPU support for Keras and TensorFlow
Checking for a CUDA-capable NVIDIA GPU
https://medium.com/data-science-365/setting-up-a-deep-learning-workplace-with-an-nvidia-graphics-card-gpu-for-windows-os-b6bff06eeec7 6/25
06/07/23, 13:16 Setting up a Deep Learning Workplace with an NVIDIA Graphics Card (GPU) — for Windows OS | by Rukshan Pramoditha | D…

CUDA (by NVIDIA) is the most popular platform for parallel computing and
programming. Almost all the GPU-supported Python libraries like CatBoost,
TensorFlow, Keras, PyTorch, OpenCV, and CuPy were designed to run on NVIDIA
CUDA-enabled graphics cards. That's why we’re searching for a CUDA-capable
NVIDIA GPU.

Step 1: Check whether there is any GPU and its part number if a GPU is
available. There are two ways to do it.

1. Check the specification of your device given by the manufacturer.

2. Go to Windows Device Manager → Display Adapters. In my case, I have an

NVIDIA GeForce RTX 3050 Ti Laptop GPU.

(Image by author)

Step 2: If you have an NVIDIA GPU, check for its CUDA compatibility by visiting
this website. Choose the correct section that matches your GPU part number.
For me, it is the “CUDA-Enabled GeForce and TITAN Products” section on the
website. If your GPU is listed in the table, it is CUDA-capable (Hint: Almost all
the recently-launched NVIDIA graphics cards should be CUDA capable).

If your graphics card is not CUDA capable, you do not need to follow the below
steps. Alternatively, you can use Colab’s free GPU!
Installing the GPU drivers
If you don’t have GPU drivers installed on your PC or laptop, you need to install the
correct driver through this website. If your GPU driver is already shown in the
Display Adaptors, you can skip this step.

https://medium.com/data-science-365/setting-up-a-deep-learning-workplace-with-an-nvidia-graphics-card-gpu-for-windows-os-b6bff06eeec7 7/25
06/07/23, 13:16 Setting up a Deep Learning Workplace with an NVIDIA Graphics Card (GPU) — for Windows OS | by Rukshan Pramoditha | D…

Selecting the right driver for my NVIDIA GPU (Image from NVIDIA website, screenshot by author)

Installing Microsoft Visual Studio

As I mentioned above, this is a prerequisite for CUDA Toolkit. It means that you
should install this before installing CUDA Toolkit. Go to this website and download
the Community 2022 Edition of Visual Studio. Just install the basic (default)
components and you don’t need to add any other workloads.
Installing CUDA Toolkit
Step 1: Visit this website to see the CUDA Toolkit version number that is
compatible with the installed Python and TensorFlow versions.

Selecting the right version for the CUDA Toolkit (Image from TensorFlow website, screenshot by author)

The most important thing is selecting the correct CUDA Toolkit version that
matches the installed Python and TensorFlow versions. We installed Python 3.9 and
TensorFlow 2.9.0. So, the correct CUDA Toolkit version is 11.2. At the time of writing
this tutorial, CUDA Toolkit version 11.7.0 (May 2022) is also available. You may
consider installing this newest version to get the latest features but you might get

https://medium.com/data-science-365/setting-up-a-deep-learning-workplace-with-an-nvidia-graphics-card-gpu-for-windows-os-b6bff06eeec7 8/25
06/07/23, 13:16 Setting up a Deep Learning Workplace with an NVIDIA Graphics Card (GPU) — for Windows OS | by Rukshan Pramoditha | D…

compatibility issues because the newest version has not yet been fully tested with
Python 3.9 and TensorFlow 2.9.0.

Step 2: Go to this website to download the CUDA Toolkit. Select the version,
operating system type, architecture, operating system version and installer type.

Step 3: Download and save the CUDA Toolkit software.

Step 4: Run the .exe file and follow the on-screen instructions to install the
CUDA Toolkit. The important thing is to keep the recommended settings and
default selections when you install CUDA. Only change them if you know what
you’re doing!

(Image by author)

https://medium.com/data-science-365/setting-up-a-deep-learning-workplace-with-an-nvidia-graphics-card-gpu-for-windows-os-b6bff06eeec7 9/25
06/07/23, 13:16 Setting up a Deep Learning Workplace with an NVIDIA Graphics Card (GPU) — for Windows OS | by Rukshan Pramoditha | D…

(Image by author)

(Image by author)

Installing cuDNN SDK

https://medium.com/data-science-365/setting-up-a-deep-learning-workplace-with-an-nvidia-graphics-card-gpu-for-windows-os-b6bff06eeec7 10/25
06/07/23, 13:16 Setting up a Deep Learning Workplace with an NVIDIA Graphics Card (GPU) — for Windows OS | by Rukshan Pramoditha | D…

This is not an install. It is just a copy-paste process.

Step 1: Visit this website to see the cuDNN SDK version number that is compatible
with the installed Python and TensorFlow versions.

Selecting the right version for the cuDNN SDK (Image from TensorFlow website, screenshot by author)

The most important thing is selecting the correct cuDNN SDK version that matches
the installed Python, TensorFlow and CUDA Toolkit versions. We installed Python
3.9 and TensorFlow 2.9.0. So, the correct cuDNN SDK version is 8.1. At the time of
writing this tutorial, cuDNN SDK version 8.4.0 (April 2022) is also available for CUDA
Toolkit 11.x. You may consider installing this newest version to get the latest
features but you might get compatibility issues because the newest version has not
yet been fully tested with Python 3.9 and TensorFlow 2.9.0.

Step 2: Go to this website to download the cuDNN SDK. You should have an
account on NVIDIA to proceed. You will also need to fill out a small survey form
before downloading.

Step 3: Download the .zip file.

Step 4: Extract the .zip file and copy the following folders inside it.

(Image by author)

https://medium.com/data-science-365/setting-up-a-deep-learning-workplace-with-an-nvidia-graphics-card-gpu-for-windows-os-b6bff06eeec7 11/25
06/07/23, 13:16 Setting up a Deep Learning Workplace with an NVIDIA Graphics Card (GPU) — for Windows OS | by Rukshan Pramoditha | D…

Step 5: Go the following path. Your path may vary depending on the architecture
and the version numbers of installed software. So, do not copy my path!

C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.7

Step 6: Paste the copied folders. The system may ask to replace the files.

Step 7: Go to the bin folder and copy the path. Go to the libnvvp folder and copy
the path. Your paths may vary depending on the architecture and the version
numbers of installed software. So, do not copy my paths!

C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.7\bin

C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.7\libnvvp

Step 7: Add these paths to the system’s Path environment variable.

(Image by author)

Open in app Sign up Sign In

https://medium.com/data-science-365/setting-up-a-deep-learning-workplace-with-an-nvidia-graphics-card-gpu-for-windows-os-b6bff06eeec7 12/25
06/07/23, 13:16 Setting up a Deep Learning Workplace with an NVIDIA Graphics Card (GPU) — for Windows OS | by Rukshan Pramoditha | D…

(Image by author)

(Image by author)

https://medium.com/data-science-365/setting-up-a-deep-learning-workplace-with-an-nvidia-graphics-card-gpu-for-windows-os-b6bff06eeec7 13/25
06/07/23, 13:16 Setting up a Deep Learning Workplace with an NVIDIA Graphics Card (GPU) — for Windows OS | by Rukshan Pramoditha | D…

(Image by author)

Verifying the GPU detection

After you have done all the installations, it is time to verify them. Open a Python 3
Jupyter Notebook from the Anaconda Navigator and execute one of the following
commands.

import tensorflow as tf
tf.test.gpu_device_name()

(Image by author)

import tensorflow as tf
tf.config.list_physical_devices('GPU')

https://medium.com/data-science-365/setting-up-a-deep-learning-workplace-with-an-nvidia-graphics-card-gpu-for-windows-os-b6bff06eeec7 14/25
06/07/23, 13:16 Setting up a Deep Learning Workplace with an NVIDIA Graphics Card (GPU) — for Windows OS | by Rukshan Pramoditha | D…

(Image by author)

len(tf.config.list_physical_devices('GPU'))

In my case, this returns 1 — The number of available GPUs.

Congratulations! TensorFlow has detected the NVIDIA GPU on your system. From
now on, Keras and TensorFlow will run on GPU by default.

import tensorflow as tf
tf.debugging.set_log_device_placement(True)
# Create some tensors
a = tf.constant([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]])
b = tf.constant([[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]])
c = tf.matmul(a, b)
print(c)

(Image by author)

To force TensorFlow or Keras to run on CPU, we can use with tf.device(): to create
a device context and put all the operations within that context.

import tensorflow as tf
tf.debugging.set_log_device_placement(True)
# Place tensors on the CPU
with tf.device('/device:CPU:0'):
a = tf.constant([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]])
b = tf.constant([[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]])
# Run on the GPU (default)
c = tf.matmul(a, b)

https://medium.com/data-science-365/setting-up-a-deep-learning-workplace-with-an-nvidia-graphics-card-gpu-for-windows-os-b6bff06eeec7 15/25
06/07/23, 13:16 Setting up a Deep Learning Workplace with an NVIDIA Graphics Card (GPU) — for Windows OS | by Rukshan Pramoditha | D…

(Image by author)

The tensors a and b are not run on the GPU by default. They are run on the CPU.

This is the end of today’s post.

Please let me know if you’ve any questions or feedback.

I hope you enjoyed reading this article. If you’d like to support me as a writer, kindly
consider signing up for a membership to get unlimited access to Medium. It only costs $5
per month and I will receive a portion of your membership fee.

Join Medium with my referral link - Rukshan Pramoditha

As a Medium member, a portion of your membership fee goes to
writers you read, and you get full access to every story…
rukshanpramoditha.medium.com

Thank you so much for your continuous support! See you in the next article. Happy
learning to everyone!

Read next (recommended) — Written by me!

Why GPUs for Machine Learning and Deep Learning?

Why GPUs for Machine Learning and Deep Learning?

Unlock the power of data parallelism with high-performance GPUs
rukshanpramoditha.medium.com

https://medium.com/data-science-365/setting-up-a-deep-learning-workplace-with-an-nvidia-graphics-card-gpu-for-windows-os-b6bff06eeec7 16/25
06/07/23, 13:16 Setting up a Deep Learning Workplace with an NVIDIA Graphics Card (GPU) — for Windows OS | by Rukshan Pramoditha | D…

Don’t have a laptop or PC with an NVIDIA GPU? Don’t worry! Google Colab provides
you with free GPU resources.

How to Use Google Colab Free GPU to Run Deep Learning Code
Incredibly Faster
Neural Networks and Deep Learning Course: Part 13
rukshanpramoditha.medium.com

Choose the right laptop for data science, machine learning and deep learning tasks.

20 Necessary Requirements of a Perfect Laptop for Data Science

and Machine Learning Tasks
towardsdatascience.com

Rukshan Pramoditha
2022–05–23

Deep Learning Artificial Intelligence Nvidia TensorFlow Keras

Follow

https://medium.com/data-science-365/setting-up-a-deep-learning-workplace-with-an-nvidia-graphics-card-gpu-for-windows-os-b6bff06eeec7 17/25
06/07/23, 13:16 Setting up a Deep Learning Workplace with an NVIDIA Graphics Card (GPU) — for Windows OS | by Rukshan Pramoditha | D…

Written by Rukshan Pramoditha

5.7K Followers · Editor for Data Science 365

2,000,000+ Views | BSc in Stats | Top 50 Data Science/AI/ML Writer on Medium | Sign up:
https://rukshanpramoditha.medium.com/membership

More from Rukshan Pramoditha and Data Science 365

Rukshan Pramoditha

How to Land a Data Science Job With Little or No Experience

12 valuable tips to consider

· 6 min read · Apr 28

44 1

https://medium.com/data-science-365/setting-up-a-deep-learning-workplace-with-an-nvidia-graphics-card-gpu-for-windows-os-b6bff06eeec7 18/25
06/07/23, 13:16 Setting up a Deep Learning Workplace with an NVIDIA Graphics Card (GPU) — for Windows OS | by Rukshan Pramoditha | D…

Rukshan Pramoditha in Data Science 365

4 Powerful and Affordable Laptops for Data Science, ML and DL in 2023

Best laptops for data science, machine learning and deep learning under $800. No MacBook
included! See the reasons at the end

· 8 min read · Jan 12

322 3

Rukshan Pramoditha in Data Science 365

Principal Component Analysis (PCA) with Scikit-learn

https://medium.com/data-science-365/setting-up-a-deep-learning-workplace-with-an-nvidia-graphics-card-gpu-for-windows-os-b6bff06eeec7 19/25
06/07/23, 13:16 Setting up a Deep Learning Workplace with an NVIDIA Graphics Card (GPU) — for Windows OS | by Rukshan Pramoditha | D…

Unsupervised Machine Learning Algorithm for Dimensionality Reduction

· 12 min read · Aug 4, 2020

376 1

Rukshan Pramoditha in Towards Data Science

20 Necessary Requirements of a Perfect Laptop for Data Science and

Machine Learning Tasks
· 7 min read · Jun 5, 2021

784 19

See all from Rukshan Pramoditha

See all from Data Science 365

https://medium.com/data-science-365/setting-up-a-deep-learning-workplace-with-an-nvidia-graphics-card-gpu-for-windows-os-b6bff06eeec7 20/25
06/07/23, 13:16 Setting up a Deep Learning Workplace with an NVIDIA Graphics Card (GPU) — for Windows OS | by Rukshan Pramoditha | D…

Recommended from Medium

Thai Tran ✅ in 𝐀𝐈 𝐦𝐨𝐧𝐤𝐬

Training PyTorch models on a Mac M1 and M2

PyTorch models on Apple Silicon M1 and M2

· 9 min read · Mar 24

61

https://medium.com/data-science-365/setting-up-a-deep-learning-workplace-with-an-nvidia-graphics-card-gpu-for-windows-os-b6bff06eeec7 21/25
06/07/23, 13:16 Setting up a Deep Learning Workplace with an NVIDIA Graphics Card (GPU) — for Windows OS | by Rukshan Pramoditha | D…

Rukshan Pramoditha in Data Science 365

Determining the Right Batch Size for a Neural Network to Get Better and
Faster Results
Guidelines for choosing the right batch size to maintain optimal training speed and accuracy
while saving computer resources

· 4 min read · Sep 26, 2022

38

Lists

AI Regulation
6 stories · 16 saves

ChatGPT
21 stories · 23 saves

ChatGPT prompts
17 stories · 62 saves

Natural Language Processing

373 stories · 29 saves

https://medium.com/data-science-365/setting-up-a-deep-learning-workplace-with-an-nvidia-graphics-card-gpu-for-windows-os-b6bff06eeec7 22/25
06/07/23, 13:16 Setting up a Deep Learning Workplace with an NVIDIA Graphics Card (GPU) — for Windows OS | by Rukshan Pramoditha | D…

Clément Delteil in Towards AI

Creating Stunning Neural Network Visualizations with ChatGPT and

PlotNeuralNet
Presenting PlotNeuralNet, a LaTeX / Python package to visualize Neural Networks

· 8 min read · Mar 8

105

Nikos Kafritsas in Towards Data Science

https://medium.com/data-science-365/setting-up-a-deep-learning-workplace-with-an-nvidia-graphics-card-gpu-for-windows-os-b6bff06eeec7 23/25
06/07/23, 13:16 Setting up a Deep Learning Workplace with an NVIDIA Graphics Card (GPU) — for Windows OS | by Rukshan Pramoditha | D…

Temporal Fusion Transformer: Time Series Forecasting with Deep

Learning — Complete Tutorial
Create accurate & interpretable predictions

· 12 min read · Nov 5, 2022

1.5K 18

Rukshan Pramoditha in Towards Data Science

Convolutional vs Feedforward Autoencoders for Image Denoising

Cleaning corrupted images using convolutional and feedforward autoencoders

· 9 min read · Jan 24

62

https://medium.com/data-science-365/setting-up-a-deep-learning-workplace-with-an-nvidia-graphics-card-gpu-for-windows-os-b6bff06eeec7 24/25
06/07/23, 13:16 Setting up a Deep Learning Workplace with an NVIDIA Graphics Card (GPU) — for Windows OS | by Rukshan Pramoditha | D…

Rukshan Pramoditha in Data Science 365

4 Powerful and Affordable Laptops for Data Science, ML and DL in 2023

Best laptops for data science, machine learning and deep learning under $800. No MacBook
included! See the reasons at the end

· 8 min read · Jan 12

322 3

See more recommendations

https://medium.com/data-science-365/setting-up-a-deep-learning-workplace-with-an-nvidia-graphics-card-gpu-for-windows-os-b6bff06eeec7 25/25