torch.cuda.is_available() always return false

[knwin]

I have installed segment geospatial as instructed on the site. OS is windows 11 with Nvidia RTX 4050 (6GB).
It seems it does not use GPU as it always return False from torch.cuda.is_available().
How do I resolve it?

[giswqs]

Check out this blog post.

A summary of the steps:

• Step 1: Install Microsoft Visual Studio
  • Download and install Microsoft visual studio community version. Select .NET desktop development and Desktop development with C++
• Step 2: Install Nvidia CUDA Toolkit
  • http://developer.nvidia.com/cuda-downloads
• Test cuda with nvidia-smi

[knwin]

Thanks Sir,
I have failed after many tries. Torch modules is always cpu version when install via mamba/conda.
So I make another installation with pip. I additionally installed cuda verson of torch as follow.. (ref: here)
pip3 install --ignore-installed torch==2.0.1 torchvision==0.15.2 --index-url https://download.pytorch.org/whl/cu117

Then torch.cuda.is_available() return true.

( I ran into another problem while running sample note book :( )

Thanks for your support.

[oscarbau]

I faced the same issue, you need to be really careful with the versions of Microsoft Visual Studio, Nvidia CUDA Toolkit, GPU model, and Python in your environment. They all need to be compatible with each other

[knwin]

I found multiple Microsoft Visual Studio in my laptop. 2010,2012 (x64 & x86), 2013 (x64 & x86), 2015-2022 (x64 & x86). Is it ok to have multiple versions?

[oscarbau]

I don't know, you need to test.