[SOLVED] PyTorch installation process - questions

Hello guys, i am new to PyTorch and i have some questions related to the installation process.

So far, i have done these:

  1. Installed Ubuntu 18.04, 2. Updated python to 3.7, 3. Installed Anaconda, 4. Made Ubuntu use my GPU instead of Intel Graphics

Now, correct me if i am wrong, i need to install Cuda before proceeding to the installation of PyTorch. PyTorch doesn’t work with Cuda 10 yet, but i can’t download Cuda 9.2 or 8 for the Ubuntu 18.04. What should i do?

Also, my GPU is kinda bad (nVidia GEFORCE GT630M). Is it compatible with PyTorch? If not, what can i do?

Thanks in advance

Hi,

Your GPU, GT630M has compute capability of 2.1. Unfortunately, this is quite old (6years) and not supported by pytorch.
Anyway, any recent CPU will actually be faster than such and old GPU, so you can just use your CPU. To do so, forget about cuda versions and just install pytorch with no cuda support :slight_smile:

Ok, thank you for the answer! However, i will be able to train models on the cloud if i want, regardless of the installation setup i choose, right?

Yes, you will need to setup your cloud machine with cuda support if you have gpus there.
The change to the code to support GPU is minimal as it’s just a few .cuda() calls, also you can use torch.cuda.is_available() to check if it is available on the current machine.

Thank you very much! One last question :slight_smile: Do you know which cloud service (like AWS, Azure etc) is the best regarding the ease of use?

I don’t use them so I couldn’t tell you what I personally prefer.
They all provide pre-built VM iirc so pick the one your prefer/already use !

Ok man, thank you for the clarifications!