pytutorial/pytorch/layers/cuda

Cuda

{:.no_toc}

* TOC {:toc}

The goal

Convince PyTorch and Nvidia's GPUs working together. I assume you installed the PyTorch and/or TensorFlow version for CUDA. (see Python installation instructions on thsi site...)

Questions to David Rotermund

Windows

• Download and install CUDA driver
• Download and install cuDNN toolkit (you will need to create an account :-( )

Test the PyTorch

import torch
torch.cuda.is_available()

Expected output:

True

torch.cuda.is_available()

Expected output:

True

torch.backends.cuda.is_built()

Expected output:

True

torch.backends.cudnn.version()

Expected output (number depends on the GPU generation and may be different):

8904

torch.backends.cudnn.enabled

Expected output:

True

my_cuda_device = torch.device('cuda:0')
print(torch.cuda.get_device_properties(my_cuda_device))

Expected output (values depend on the GPU generation):

_CudaDeviceProperties(name='NVIDIA GeForce RTX 3060', major=8, minor=6, total_memory=12011MB, multi_processor_count=28)