How to make code run on GPU on Windows 10?

Question

I want to run my code run on gpu in windows 10, like for google colab, we can just change the runtime option which is pretty easy to do to shift to gpu. Is there a possibility to do the same for jupyter notebook in windows.

I see. Regardless, Jupyter is running a Python kernel, so any code you write in it should run the same as on the host itself, so saying "running jupyter on gpu" isn't accurate — OneCricketeer, May 17 '21 at 11:39
@OneCricketeer Yes it is running a python kernel as you said that's why it takes endless days to create a deep learning model. For the same above reason I wanted jupyter to run on gpu to decrease computing time. — RShiny_Noob, May 24 '21 at 06:07
Like I said "jupyter" isn't what's running your gpu code, the kernel is, which would be IPython — OneCricketeer, May 25 '21 at 11:49
@OneCricketeer Okay so is there a way to run IPython on local gpu? — RShiny_Noob, May 26 '21 at 15:16
The below answer should work. If you're trying to use pytorch, maybe you can share the steps you've tried so far? And what isn't working? — OneCricketeer, May 27 '21 at 12:08
@OneCricketeer In the below answer, I got stuck at step 7, that import statement is not working. — RShiny_Noob, May 29 '21 at 17:55
As of tensorflow 2, you wouldn't use `tensorflow-gpu`. https://www.tensorflow.org/install/pip But I thought you wanted pytorch anyway? https://stackoverflow.com/questions/48152674/how-to-check-if-pytorch-is-using-the-gpu — OneCricketeer, May 30 '21 at 13:26
You can check the following answer https://stackoverflow.com/questions/51002045/how-to-make-jupyter-notebook-to-run-on-gpu — E.Zolduoarrati, Dec 03 '21 at 08:18

score 1 · Answer 1 · edited May 13 '21 at 14:56

1

You will actually need to use tensorflow-gpu to run your jupyter notebook on a gpu.

The best way to achieve this would be

Install Anaconda on your system
Download cuDNN & Cuda Toolkit 11.3 .
Add cuDNN and Cuda Toolkit to your PATH.
Create an environment in Anaconda
pip install tensorflow-gpu
pip install [jupyter-notebook/jupyterlab]
Import tensorflow-gpu in your notebook
Enjoy. You can now run your notebook on your GPU

edited May 13 '21 at 14:56

Dharman

30,962
25
85
135

answered May 13 '21 at 14:51

Veer Shah

11
1

Hey, I have a nvidia 940mx (compute capability - 5.0) on my system so should I download older version of both CuDNN and Cuda Toolkit OR It will work with the latest versions? Thanks. – RShiny_Noob May 17 '21 at 05:44
Hi, I'm not able to import tensorflow-gpu, it says syntax error. Although I can use "import tensorflow" but how do I make sure that whether jupyter is using gpu or not. – RShiny_Noob May 24 '21 at 10:05

score 0 · Answer 2 · answered Mar 17 '22 at 17:44

Easy Direct way Create a new environment with TensorFlow-GPU and activate it whenever you want to run your code in GPU

Open Anaconda promote and Write

Conda create --name tf_GPU tensorFlow-gpu

Now it's time to test if our code Run on GPU or CPU

Conda activate tf_GPU --- (Activating the env)
Jupyter notebook ----(Open notebook from the tf_GPU env)

if this Code gives you 1 this means you are runing on GPU

print("Num GPUs Available: ", len(tf.config.list_physical_devices('GPU')))

You can also use this code to make sure you run on GPU

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)

Geting output like this means you using CPU not GPU

Executing op _EagerConst in device /job:localhost/replica:0/task:0/device:CPU:0
Executing op _EagerConst in device /job:localhost/replica:0/task:0/device:CPU:0
Executing op MatMul in device /job:localhost/replica:0/task:0/device:CPU:0
tf.Tensor(
[[22. 28.]
 [49. 64.]], shape=(2, 2), dtype=float32)

How to make code run on GPU on Windows 10?

2 Answers2

Easy Direct way Create a new environment with TensorFlow-GPU and activate it whenever you want to run your code in GPU