Cuda isn't Available

1

Hello. I am using a machine on cloud with two Nvidia Tesla A100 and have alredy installed pytorch, but it seems to still does not recognize cuda afterall.

I have installed the latest pytorch version with

pip3 install torch==1.10.0+cu113 torchvision==0.11.1+cu113 torchaudio==0.10.0+cu113 -f https://download.pytorch.org/whl/cu113/torch_stable.html

Importing torch and asking for torch.version.cuda I get ‘11.3’.

Since it is a system with multiple users, I gess paths may be different from default, but what procedure should I take to resolve this?

System has the following software installed

CUDA Toolkit 11.4
nvhpc-21.7 - NVIDIA HPC SDK
NVIDIA cuDNN

2

To use the pip wheels you would only need to install an appropriate NVIDIA driver on your workstation so I guess that your driver installation is either not working or your environment is not supporting GPUs (e.g. if you are launching a docker container without the --gpus all flag).