Hello, I’m using Mobaxterm to access a server to work on training a model when I try to run the command to launch it, I have his error:
NVIDIA GeForce RTX 3090 with CUDA capability sm_86 is not compatible with the current PyTorch installation.
The current PyTorch install supports CUDA capabilities sm_37 sm_50 sm_60 sm_61 sm_70 sm_75 compute_37.
If you want to use the NVIDIA GeForce RTX 3090 GPU with PyTorch, please check the instructions at Start Locally | PyTorch
Unfortunately, I have no clue how to use Pytorch, I already used the command from Pytorch website to install the latest version but well it doesn’t seem to work, can anyone help me, I’ve been stuck for a few days?? I used this command to install it: conda install pytorch torchvision torchaudio pytorch-cuda=12.1 -c pytorch -c nvidia
note: I have absolutely no idea how pytorch works and i’m not that good in programming, I looked for a solution before posting but it was confusing and didn’t have a similar case
Thank you.
You have most likely installed a PyTorch binary with CUDA 10.2 dependencies, which should also be visible via torch.version.cuda. Uninstall it and reinstall the right one with CUDA 11.8 or 12.1.
I’m sorry that i still don’t get it. With this command conda install pytorch torchvision torchaudio pytorch-cuda=12.1 -c pytorch -c nvidia i’m still not getting the latest pytorch? (Is this even how to get binaries??)
Also is this a command that I can run? Because it’s not working for me
You should, but as you can see in related posts usually the user’s environment is using multiple and conflicting PyTorch installations.
It works for me:
torch.version.cuda
Out[2]: '12.1'
If this does not return anything or raises an error, your installation might indeed be corrupt.
I would then recommend to create a new and empty virtual environment and to install the desired PyTorch binary only once.