When is all sm_120 supported in nightly?

1

When is all sm_120 supported in nightly?

I switched to gigabyte 5060ti graphics card on a 1660 and installed the driver properly. Since then, automatic1111 hasn’t run. So I deleted the existing pytorch118 and reinstalled it with nightly pytorch128. So pytorch only has 1 binary.

However, the same problem arises.
The statement does not support sm_120.
Is it possible that some sm_120s are not supported?
When is all sm_120 supported in nightly?

NVIDIA GeForce RTX 5060 Ti with CUDA capability sm_120 is not compatible with the current PyTorch installation. The current PyTorch install supports CUDA capabilities sm_50 sm_60 sm_61 sm_70 sm_75 sm_80 sm_86 sm_90. If you want to use the NVIDIA GeForce RTX 5060 Ti GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/ C:\Users\나현준>pip list | findstr torch torch 2.8.0.dev20250513+cu128 torchaudio 2.6.0.dev20250514+cu128 torchvision 0.22.0.dev20250514+cu128

C:\Users\Na Hyunjun>
C:\Users\나현준>pip show torch
Name: torch
Version: 2.8.0.dev20250513+cu128
Summary: Tensors and Dynamic neural networks in Python with strong GPU acceleration
Home-page: https://pytorch.org/
Author: PyTorch Team
Author-email: packages@pytorch.org
License: BSD-3-Clause
Location: c:\users\나현준\appdata\local\programs\python\python310\lib\site-packages
Requires: filelock, fsspec, jinja2, networkx, sympy, typing-extensions
Required-by: torchaudio, torchvision

C:\Users\나현준>pip show torchvision
Name: torchvision
Version: 0.22.0.dev20250514+cu128
Summary: image and video datasets and models for torch deep learning
Home-page: GitHub - pytorch/vision: Datasets, Transforms and Models specific to Computer Vision
Author: PyTorch Core Team
Author-email: soumith@pytorch.org
License: BSD
Location: c:\users\나현준\appdata\local\programs\python\python310\lib\site-packages
Requires: numpy, pillow, torch
Required-by:

C:\Users\나현준>pip show torchaudio
Name: torchaudio
Version: 2.6.0.dev20250514+cu128
Summary: An audio package for PyTorch
Home-page: GitHub - pytorch/audio: Data manipulation and transformation for audio signal processing, powered by PyTorch
Author: Soumith Chintala, David Pollack, Sean Naren, Peter Goldsborough, Moto Hira, Caroline Chen, Jeff Hwang, Zhaoheng Ni, Xiaohui Zhang
Author-email: soumith@pytorch.org
License:
Location: c:\users\나현준\appdata\local\programs\python\python310\lib\site-packages
Requires: torch
Required-by:

<Check if it’s viable>
Python 3.10.6 (tags/v3.10.6:9c7b4bd, Aug 1 2022, 21:53:49) [MSC v.1932 64 bit (AMD64)] on win32
Type “help”, “copyright”, “credits” or “license” for more information.

import torch
print(torch.version.cuda)
12.8
print(torch.cuda.get_device_capability())
(12, 0)
exit()

2

All PyTorch binaries with CUDA 12.8 support sm_120 and the earliest were released for Linux in February. Based on the error message you are using an older binary built with an older CUDA toolkit. I would recommend executing a simple test after installing the right binary via:

print(torch.__version__)
print(torch.cuda.version)
print(torch.cuda.get_arch_list())
print(torch.cuda.get_device_properties(0))
print(torch.randn(1).cuda())

Execute this code snippet standalone without any other application and post the output here.

3

image
image1137×156 5.16 KB

under code is starting page of automatic1111

image
image957×286 8.6 KB

4

Thank you for confirming the PyTorch binary itself works fine as seen in your first screenshot!
The output shows CUDA 12.8 is used and the binaries support compute capabilities 6.1 to 12.0. The last line of code also shows you are able to use your Blackwell GPU by creating a random tensor on the GPU.

I’m not familiar with automatic1111 and guess it’s using another virtual environment with another (potentially older) PyTorch binary.