PyTorch version for cuda compute capability 3.0 (GTX 780M)

I installed PyTorch on my laptop only to find out that my GPU is not supported. Is there a version that I can use that still supports my GPU ? Or are there any other solutions ?

/home/haziq/anaconda3/envs/dl/lib/python3.5/site-packages/torch/cuda/init.py:97: UserWarning:
Found GPU0 GeForce GTX 780M which is of cuda capability 3.0.
PyTorch no longer supports this GPU because it is too old.

warnings.warn(old_gpu_warn % (d, name, major, capability[1]))

You could build PyTorch from source.
The installation is described here.
Let me know, if you encounter any problems.

1 Like

I saw this in another thread PyTorch no longer supports this GPU because it is too old.

We’ve got PyTorch CUDA bundles 167 with compute capabilities starting 3.0, 3.5
There is no separate python package, but you can extract the package from the installer archive.

which linked to this webpage http://jetware.io/appliances/jetware/pytorch03_python3_cuda91#appliance_usage_vm_images. I saw a VirtualBox image download and so was wondering if this is an alternative ? Would prefer not to build from source as I am really good at encountering problems.

Oh, I haven’t used it, so I cannot comment on the jetware images.
Haha, you should give it a try. It’s quite simple. Just follow the steps from my previous link and try it out. :wink:

Answer to original question: PyTorch 0.3.0 is the last release to support compute capability 3.0

source: https://github.com/pytorch/pytorch/releases/tag/v0.3.1

Removed support for CUDA capability 3.0 and 5.0

1 Like

I am new to this and I am using gtx 770 i need to train Yolo models and I really need help regarding this PyTorch from source thing. Can you please guide me in this ? should i do exact same steps mentioned in this ?

Yes, read the instructions from my link and follow them. I would also recommend to stick to a clean environment (e.g. create a new virtual conda environment) and also a “common” OS such as Ubuntu 20.04. Other linux flavors would of course also work, but you might need to install the appropriate toolchain.

1 Like

Oh, so is Linux or Ubuntu necessary? As I am using windows

No, Linux is not necessary and you should be able to build PyTorch using Windows. However, I’m not deeply familiar with Windows and its build process/setup.

Alright, So i shall try followinf your links points of windows and will let you know if theres any problem occurs

OK, I am following the steps but I have some confusion if you can help.

  1. I am using Cuda 11.6 with VS code 2022. for my gtx 770 to work
  2. Its mentioned the legacy version (I don’t know whether I should do it as I don’t know what it is or should I just go with the same page till the end)
  3. Do I have to follow Cuda Build? For GPU to work
  1. I don’t know what “legacy version” refers to, but am also not familiar with Windows builds.
  2. Yes, you need to use the CUDA compiler to build PyTorch with GPU support.

OK Perfect will do that

Hi, it has been a long time though as I thought of buying a new GPU but still it will take time to deliver so I am back again :face_with_diagonal_mouth: Anyways I have implemented each and every step and opened Jupiter notebook but it showed the error as “Torch not compiled with CUDA enabled”

Hi did it work? I have been into this for a long time still cant able to do it.

GeForce GTX 770 is a very old architecture (released in 2013), I don’t think it’s supported with recent PyTorch.