Install Tensorflow-gpu 2.4.0 with Cuda 11.0 and CuDnn 8 Using Anaconda

Photo by Christian Wiediger on Unsplash

Have you been frustrated, installing Tensorflow Gpu with Cuda and all stuff; If yes, This Blog is for you, here you’ll get an easy way to install Tensorflow GPU with the latest versions.CUDA is a parallel computing platform and programming model developed by NVIDIA for general computing on graphical processing units (GPUs). With CUDA, developers can dramatically speed up computing applications by harnessing the power of GPUs.

The CUDA Toolkit from NVIDIA provides everything you need to develop GPU-accelerated applications. This CUDA Toolkit includes GPU-accelerated libraries and the CUDA runtime for the Conda ecosystem. For the full CUDA Toolkit with a compiler and development tools visit https://developer.nvidia.com/cuda-downloads

License Agreements The packages are governed by the CUDA Toolkit End User License Agreement (EULA). By downloading and using the packages, you accept the terms and conditions of the CUDA EULA — https://docs.nvidia.com/cuda/eula/index.html

Here is the Version list of all the Libraries:

tensorflow-gpu==2.4.0

cudatoolkit==11.0

cuDnn==8

python==3.7 (or later)

We’ll be following 6 steps in order to install, tensorflow-gpu version 2.4 successfully.

1. First of all Download Cuda 11.0 compactable, CuDnn version 8 from Nvidia’s official website here. Then extract and Keep it aside, the files should look like this,

Downloaded CuDnn 8 should have these files.

2. Create a new Conda environment with python 3.7 or later,

conda create -n myenv python=3.7

run the above code to create a new environment with python 3.7.

3. Here comes the main part, now we need to install Cuda toolkit, You can download it from Nvidia’s official website or directly using Anaconda prompt in 2 steps,

conda activate <env>
conda install cudatoolkit

Official Conda webite

Just running the above code will install Cuda 11.0 within the environment and make us happy.

4. TensorFlow is an open-source software library for high-performance numerical computation. Its flexible architecture allows easy deployment of computation across a variety of platforms (CPUs, GPUs, TPUs), and from desktops to clusters of servers to mobile and edge devices.

Originally developed by researchers and engineers from the Google Brain team within Google’s AI organization, it comes with strong support for machine learning and deep learning and the flexible numerical computation core is used across many other scientific domains. Now it’s time for the installation of Tensorflow; the latest TensorFlow version is 2.4 and we need not install TensorFlow cause, tensorflow-gpu includes all. Python makes it even easier for us,

pip install tensorflow-gpu

The above command installs Tensorflow gpu version, Tensorflow estimator, Tensorflow base. Don’t use conda here cause, it’ll install Cuda 10.2 and cuDnn 7 along with that, so it may conflict with the new version installed.

The majority of the bugs, particularly in ML comes from Version confliction; it is the worst thing actually.

5. Now copy all the files from bin folder of the downloaded, cuDnn 8 folder. Then paste it in the bin folder of the conda environment folder, usually you could find the path from user in C,

C:\Users\<name>\anaconda3\envs\<env name>\Library\bin

Paste the DLL files here, and, That’s it! you made it!. Now you are ready for the GPU revolution.

Though the price of a nice performing GPU is still high, you could use the online cloud platform to render training usually they are faster than Nvidia’s MX series like Google Colaboratotry. If you have the latest GPU version, like GeForce RTX 3060 Ti or Titan series you could use the steps mentioned above to utilize the GPU.

Actually, this is my first blog, and I’m so excited to get feedback from you all, follow me on Linkedin and Github to collaborate with me. See you in the Blog,

Thank you,

Suriya