I have already briefed about tensorflow in my old blogs, in short it is an open-source library with is capable of running machine learning algorithms. Now lets jump to the installation part.
*Installation of Tensorflow with GPU support and anaconda*
Hardware requirements
NVIDIA GPU card with CUDA® Compute Capability 3.5 or higher, CUDA-enabled GPU cards
Software requirements
The following NVIDIA software must be installed on your system:
-
-
- Visual Studio is required, so download the latest version of Visual studio code
- NVIDIA® GPU drivers —CUDA 10.0 requires 410.x or higher.
- Download your driver accordingly matching your GPU (Nearly 400MB)
- CUDA® Toolkit —TensorFlow supports CUDA 10.0 (TensorFlow >= 1.13.0)
-
- Choose the latest version (CUDA Toolkit 10.1)
- Choose Platform as windows and download for your OS
- Choose Installer Type as exe(local)
- Click Download (Base Installer). It is going to take 1.3gb space.
- After Downloading, open start and type
environment variable
- In the Pop-Up box click
environment variable
- In the
system variable
locate PATH and choose edit after selecting PATH - Select NEW (after you choose the edit option for PATH)
- In the box presented to you after clicking new type
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\bin C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\libnvvp
-
- cuDNN SDK (>= 7.4.1)
- To download this you need to have an account so sign up by clicking join
- After making an account, Agree to all terms
- Download cuDNN v7.6.0 (Latest on your screen)
- Download according to your operating system(around 200MB)
- After downloading the zip file, unzip and keep it in your C drive
- Now open
environment variable
by typing it in your start menu - In the Pop-Up box click
environment variable
- In the
system variable
locate PATH and choose edit after selecting PATH - Select NEW (after you choose the edit option for PATH)
C:\cuda\bin
Add this to you NEW environment variable
- (Optional) TensorRT 5.0 to improve latency and throughput for inference on some models.
-
NOTE: You can skip the last two step if you want as its not necessary.
Installing with Anaconda
-
-
- Install Anaconda from there.
- After the installation, open a Command Prompt and type
conda create -n tensorflow
- After this gets over we can now activate tensorflow by typing
activate tensorflow
- Run the following command to install it completely
-
pip install tensorflow-gpu
-
-
We are in the ENDGAME now
To check everything works fine open python interpreter and type import tensorflow
.
import tensorflow as tf
a = tf.constant(50,name="a")
b = tf.constant(25,name="b")
c = tf.add(a,b,name="c")
sess = tf.Session()
output = sess.run(c);
print(output)
sess.close()
If it compiles without any errors VOILA!
NOTE: If you face errors then you need to search them on StackOverflow.